Loading...
1/*
2 * blkfront.c
3 *
4 * XenLinux virtual block device driver.
5 *
6 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8 * Copyright (c) 2004, Christian Limpach
9 * Copyright (c) 2004, Andrew Warfield
10 * Copyright (c) 2005, Christopher Clark
11 * Copyright (c) 2005, XenSource Ltd
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License version 2
15 * as published by the Free Software Foundation; or, when distributed
16 * separately from the Linux kernel or incorporated into other
17 * software packages, subject to the following license:
18 *
19 * Permission is hereby granted, free of charge, to any person obtaining a copy
20 * of this source file (the "Software"), to deal in the Software without
21 * restriction, including without limitation the rights to use, copy, modify,
22 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23 * and to permit persons to whom the Software is furnished to do so, subject to
24 * the following conditions:
25 *
26 * The above copyright notice and this permission notice shall be included in
27 * all copies or substantial portions of the Software.
28 *
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
35 * IN THE SOFTWARE.
36 */
37
38#include <linux/interrupt.h>
39#include <linux/blkdev.h>
40#include <linux/hdreg.h>
41#include <linux/cdrom.h>
42#include <linux/module.h>
43#include <linux/slab.h>
44#include <linux/mutex.h>
45#include <linux/scatterlist.h>
46
47#include <xen/xen.h>
48#include <xen/xenbus.h>
49#include <xen/grant_table.h>
50#include <xen/events.h>
51#include <xen/page.h>
52#include <xen/platform_pci.h>
53
54#include <xen/interface/grant_table.h>
55#include <xen/interface/io/blkif.h>
56#include <xen/interface/io/protocols.h>
57
58#include <asm/xen/hypervisor.h>
59
60enum blkif_state {
61 BLKIF_STATE_DISCONNECTED,
62 BLKIF_STATE_CONNECTED,
63 BLKIF_STATE_SUSPENDED,
64};
65
66struct blk_shadow {
67 struct blkif_request req;
68 struct request *request;
69 unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
70};
71
72static DEFINE_MUTEX(blkfront_mutex);
73static const struct block_device_operations xlvbd_block_fops;
74
75#define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
76
77/*
78 * We have one of these per vbd, whether ide, scsi or 'other'. They
79 * hang in private_data off the gendisk structure. We may end up
80 * putting all kinds of interesting stuff here :-)
81 */
82struct blkfront_info
83{
84 struct mutex mutex;
85 struct xenbus_device *xbdev;
86 struct gendisk *gd;
87 int vdevice;
88 blkif_vdev_t handle;
89 enum blkif_state connected;
90 int ring_ref;
91 struct blkif_front_ring ring;
92 struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
93 unsigned int evtchn, irq;
94 struct request_queue *rq;
95 struct work_struct work;
96 struct gnttab_free_callback callback;
97 struct blk_shadow shadow[BLK_RING_SIZE];
98 unsigned long shadow_free;
99 unsigned int feature_flush;
100 unsigned int flush_op;
101 int is_ready;
102};
103
104static DEFINE_SPINLOCK(blkif_io_lock);
105
106static unsigned int nr_minors;
107static unsigned long *minors;
108static DEFINE_SPINLOCK(minor_lock);
109
110#define MAXIMUM_OUTSTANDING_BLOCK_REQS \
111 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
112#define GRANT_INVALID_REF 0
113
114#define PARTS_PER_DISK 16
115#define PARTS_PER_EXT_DISK 256
116
117#define BLKIF_MAJOR(dev) ((dev)>>8)
118#define BLKIF_MINOR(dev) ((dev) & 0xff)
119
120#define EXT_SHIFT 28
121#define EXTENDED (1<<EXT_SHIFT)
122#define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
123#define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
124#define EMULATED_HD_DISK_MINOR_OFFSET (0)
125#define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
126#define EMULATED_SD_DISK_MINOR_OFFSET (0)
127#define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
128
129#define DEV_NAME "xvd" /* name in /dev */
130
131static int get_id_from_freelist(struct blkfront_info *info)
132{
133 unsigned long free = info->shadow_free;
134 BUG_ON(free >= BLK_RING_SIZE);
135 info->shadow_free = info->shadow[free].req.id;
136 info->shadow[free].req.id = 0x0fffffee; /* debug */
137 return free;
138}
139
140static void add_id_to_freelist(struct blkfront_info *info,
141 unsigned long id)
142{
143 info->shadow[id].req.id = info->shadow_free;
144 info->shadow[id].request = NULL;
145 info->shadow_free = id;
146}
147
148static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
149{
150 unsigned int end = minor + nr;
151 int rc;
152
153 if (end > nr_minors) {
154 unsigned long *bitmap, *old;
155
156 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap),
157 GFP_KERNEL);
158 if (bitmap == NULL)
159 return -ENOMEM;
160
161 spin_lock(&minor_lock);
162 if (end > nr_minors) {
163 old = minors;
164 memcpy(bitmap, minors,
165 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
166 minors = bitmap;
167 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
168 } else
169 old = bitmap;
170 spin_unlock(&minor_lock);
171 kfree(old);
172 }
173
174 spin_lock(&minor_lock);
175 if (find_next_bit(minors, end, minor) >= end) {
176 for (; minor < end; ++minor)
177 __set_bit(minor, minors);
178 rc = 0;
179 } else
180 rc = -EBUSY;
181 spin_unlock(&minor_lock);
182
183 return rc;
184}
185
186static void xlbd_release_minors(unsigned int minor, unsigned int nr)
187{
188 unsigned int end = minor + nr;
189
190 BUG_ON(end > nr_minors);
191 spin_lock(&minor_lock);
192 for (; minor < end; ++minor)
193 __clear_bit(minor, minors);
194 spin_unlock(&minor_lock);
195}
196
197static void blkif_restart_queue_callback(void *arg)
198{
199 struct blkfront_info *info = (struct blkfront_info *)arg;
200 schedule_work(&info->work);
201}
202
203static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
204{
205 /* We don't have real geometry info, but let's at least return
206 values consistent with the size of the device */
207 sector_t nsect = get_capacity(bd->bd_disk);
208 sector_t cylinders = nsect;
209
210 hg->heads = 0xff;
211 hg->sectors = 0x3f;
212 sector_div(cylinders, hg->heads * hg->sectors);
213 hg->cylinders = cylinders;
214 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
215 hg->cylinders = 0xffff;
216 return 0;
217}
218
219static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
220 unsigned command, unsigned long argument)
221{
222 struct blkfront_info *info = bdev->bd_disk->private_data;
223 int i;
224
225 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
226 command, (long)argument);
227
228 switch (command) {
229 case CDROMMULTISESSION:
230 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
231 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
232 if (put_user(0, (char __user *)(argument + i)))
233 return -EFAULT;
234 return 0;
235
236 case CDROM_GET_CAPABILITY: {
237 struct gendisk *gd = info->gd;
238 if (gd->flags & GENHD_FL_CD)
239 return 0;
240 return -EINVAL;
241 }
242
243 default:
244 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
245 command);*/
246 return -EINVAL; /* same return as native Linux */
247 }
248
249 return 0;
250}
251
252/*
253 * Generate a Xen blkfront IO request from a blk layer request. Reads
254 * and writes are handled as expected.
255 *
256 * @req: a request struct
257 */
258static int blkif_queue_request(struct request *req)
259{
260 struct blkfront_info *info = req->rq_disk->private_data;
261 unsigned long buffer_mfn;
262 struct blkif_request *ring_req;
263 unsigned long id;
264 unsigned int fsect, lsect;
265 int i, ref;
266 grant_ref_t gref_head;
267 struct scatterlist *sg;
268
269 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
270 return 1;
271
272 if (gnttab_alloc_grant_references(
273 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
274 gnttab_request_free_callback(
275 &info->callback,
276 blkif_restart_queue_callback,
277 info,
278 BLKIF_MAX_SEGMENTS_PER_REQUEST);
279 return 1;
280 }
281
282 /* Fill out a communications ring structure. */
283 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
284 id = get_id_from_freelist(info);
285 info->shadow[id].request = req;
286
287 ring_req->id = id;
288 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
289 ring_req->handle = info->handle;
290
291 ring_req->operation = rq_data_dir(req) ?
292 BLKIF_OP_WRITE : BLKIF_OP_READ;
293
294 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
295 /*
296 * Ideally we can do an unordered flush-to-disk. In case the
297 * backend onlysupports barriers, use that. A barrier request
298 * a superset of FUA, so we can implement it the same
299 * way. (It's also a FLUSH+FUA, since it is
300 * guaranteed ordered WRT previous writes.)
301 */
302 ring_req->operation = info->flush_op;
303 }
304
305 ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
306 BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
307
308 for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
309 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
310 fsect = sg->offset >> 9;
311 lsect = fsect + (sg->length >> 9) - 1;
312 /* install a grant reference. */
313 ref = gnttab_claim_grant_reference(&gref_head);
314 BUG_ON(ref == -ENOSPC);
315
316 gnttab_grant_foreign_access_ref(
317 ref,
318 info->xbdev->otherend_id,
319 buffer_mfn,
320 rq_data_dir(req) );
321
322 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
323 ring_req->u.rw.seg[i] =
324 (struct blkif_request_segment) {
325 .gref = ref,
326 .first_sect = fsect,
327 .last_sect = lsect };
328 }
329
330 info->ring.req_prod_pvt++;
331
332 /* Keep a private copy so we can reissue requests when recovering. */
333 info->shadow[id].req = *ring_req;
334
335 gnttab_free_grant_references(gref_head);
336
337 return 0;
338}
339
340
341static inline void flush_requests(struct blkfront_info *info)
342{
343 int notify;
344
345 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
346
347 if (notify)
348 notify_remote_via_irq(info->irq);
349}
350
351/*
352 * do_blkif_request
353 * read a block; request is in a request queue
354 */
355static void do_blkif_request(struct request_queue *rq)
356{
357 struct blkfront_info *info = NULL;
358 struct request *req;
359 int queued;
360
361 pr_debug("Entered do_blkif_request\n");
362
363 queued = 0;
364
365 while ((req = blk_peek_request(rq)) != NULL) {
366 info = req->rq_disk->private_data;
367
368 if (RING_FULL(&info->ring))
369 goto wait;
370
371 blk_start_request(req);
372
373 if (req->cmd_type != REQ_TYPE_FS) {
374 __blk_end_request_all(req, -EIO);
375 continue;
376 }
377
378 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
379 "(%u/%u) buffer:%p [%s]\n",
380 req, req->cmd, (unsigned long)blk_rq_pos(req),
381 blk_rq_cur_sectors(req), blk_rq_sectors(req),
382 req->buffer, rq_data_dir(req) ? "write" : "read");
383
384 if (blkif_queue_request(req)) {
385 blk_requeue_request(rq, req);
386wait:
387 /* Avoid pointless unplugs. */
388 blk_stop_queue(rq);
389 break;
390 }
391
392 queued++;
393 }
394
395 if (queued != 0)
396 flush_requests(info);
397}
398
399static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
400{
401 struct request_queue *rq;
402
403 rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
404 if (rq == NULL)
405 return -1;
406
407 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
408
409 /* Hard sector size and max sectors impersonate the equiv. hardware. */
410 blk_queue_logical_block_size(rq, sector_size);
411 blk_queue_max_hw_sectors(rq, 512);
412
413 /* Each segment in a request is up to an aligned page in size. */
414 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
415 blk_queue_max_segment_size(rq, PAGE_SIZE);
416
417 /* Ensure a merged request will fit in a single I/O ring slot. */
418 blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
419
420 /* Make sure buffer addresses are sector-aligned. */
421 blk_queue_dma_alignment(rq, 511);
422
423 /* Make sure we don't use bounce buffers. */
424 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
425
426 gd->queue = rq;
427
428 return 0;
429}
430
431
432static void xlvbd_flush(struct blkfront_info *info)
433{
434 blk_queue_flush(info->rq, info->feature_flush);
435 printk(KERN_INFO "blkfront: %s: %s: %s\n",
436 info->gd->disk_name,
437 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
438 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
439 "flush diskcache" : "barrier or flush"),
440 info->feature_flush ? "enabled" : "disabled");
441}
442
443static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
444{
445 int major;
446 major = BLKIF_MAJOR(vdevice);
447 *minor = BLKIF_MINOR(vdevice);
448 switch (major) {
449 case XEN_IDE0_MAJOR:
450 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
451 *minor = ((*minor / 64) * PARTS_PER_DISK) +
452 EMULATED_HD_DISK_MINOR_OFFSET;
453 break;
454 case XEN_IDE1_MAJOR:
455 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
456 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
457 EMULATED_HD_DISK_MINOR_OFFSET;
458 break;
459 case XEN_SCSI_DISK0_MAJOR:
460 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
461 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
462 break;
463 case XEN_SCSI_DISK1_MAJOR:
464 case XEN_SCSI_DISK2_MAJOR:
465 case XEN_SCSI_DISK3_MAJOR:
466 case XEN_SCSI_DISK4_MAJOR:
467 case XEN_SCSI_DISK5_MAJOR:
468 case XEN_SCSI_DISK6_MAJOR:
469 case XEN_SCSI_DISK7_MAJOR:
470 *offset = (*minor / PARTS_PER_DISK) +
471 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
472 EMULATED_SD_DISK_NAME_OFFSET;
473 *minor = *minor +
474 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
475 EMULATED_SD_DISK_MINOR_OFFSET;
476 break;
477 case XEN_SCSI_DISK8_MAJOR:
478 case XEN_SCSI_DISK9_MAJOR:
479 case XEN_SCSI_DISK10_MAJOR:
480 case XEN_SCSI_DISK11_MAJOR:
481 case XEN_SCSI_DISK12_MAJOR:
482 case XEN_SCSI_DISK13_MAJOR:
483 case XEN_SCSI_DISK14_MAJOR:
484 case XEN_SCSI_DISK15_MAJOR:
485 *offset = (*minor / PARTS_PER_DISK) +
486 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
487 EMULATED_SD_DISK_NAME_OFFSET;
488 *minor = *minor +
489 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
490 EMULATED_SD_DISK_MINOR_OFFSET;
491 break;
492 case XENVBD_MAJOR:
493 *offset = *minor / PARTS_PER_DISK;
494 break;
495 default:
496 printk(KERN_WARNING "blkfront: your disk configuration is "
497 "incorrect, please use an xvd device instead\n");
498 return -ENODEV;
499 }
500 return 0;
501}
502
503static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
504 struct blkfront_info *info,
505 u16 vdisk_info, u16 sector_size)
506{
507 struct gendisk *gd;
508 int nr_minors = 1;
509 int err;
510 unsigned int offset;
511 int minor;
512 int nr_parts;
513
514 BUG_ON(info->gd != NULL);
515 BUG_ON(info->rq != NULL);
516
517 if ((info->vdevice>>EXT_SHIFT) > 1) {
518 /* this is above the extended range; something is wrong */
519 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
520 return -ENODEV;
521 }
522
523 if (!VDEV_IS_EXTENDED(info->vdevice)) {
524 err = xen_translate_vdev(info->vdevice, &minor, &offset);
525 if (err)
526 return err;
527 nr_parts = PARTS_PER_DISK;
528 } else {
529 minor = BLKIF_MINOR_EXT(info->vdevice);
530 nr_parts = PARTS_PER_EXT_DISK;
531 offset = minor / nr_parts;
532 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
533 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
534 "emulated IDE disks,\n\t choose an xvd device name"
535 "from xvde on\n", info->vdevice);
536 }
537 err = -ENODEV;
538
539 if ((minor % nr_parts) == 0)
540 nr_minors = nr_parts;
541
542 err = xlbd_reserve_minors(minor, nr_minors);
543 if (err)
544 goto out;
545 err = -ENODEV;
546
547 gd = alloc_disk(nr_minors);
548 if (gd == NULL)
549 goto release;
550
551 if (nr_minors > 1) {
552 if (offset < 26)
553 sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
554 else
555 sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
556 'a' + ((offset / 26)-1), 'a' + (offset % 26));
557 } else {
558 if (offset < 26)
559 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
560 'a' + offset,
561 minor & (nr_parts - 1));
562 else
563 sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
564 'a' + ((offset / 26) - 1),
565 'a' + (offset % 26),
566 minor & (nr_parts - 1));
567 }
568
569 gd->major = XENVBD_MAJOR;
570 gd->first_minor = minor;
571 gd->fops = &xlvbd_block_fops;
572 gd->private_data = info;
573 gd->driverfs_dev = &(info->xbdev->dev);
574 set_capacity(gd, capacity);
575
576 if (xlvbd_init_blk_queue(gd, sector_size)) {
577 del_gendisk(gd);
578 goto release;
579 }
580
581 info->rq = gd->queue;
582 info->gd = gd;
583
584 xlvbd_flush(info);
585
586 if (vdisk_info & VDISK_READONLY)
587 set_disk_ro(gd, 1);
588
589 if (vdisk_info & VDISK_REMOVABLE)
590 gd->flags |= GENHD_FL_REMOVABLE;
591
592 if (vdisk_info & VDISK_CDROM)
593 gd->flags |= GENHD_FL_CD;
594
595 return 0;
596
597 release:
598 xlbd_release_minors(minor, nr_minors);
599 out:
600 return err;
601}
602
603static void xlvbd_release_gendisk(struct blkfront_info *info)
604{
605 unsigned int minor, nr_minors;
606 unsigned long flags;
607
608 if (info->rq == NULL)
609 return;
610
611 spin_lock_irqsave(&blkif_io_lock, flags);
612
613 /* No more blkif_request(). */
614 blk_stop_queue(info->rq);
615
616 /* No more gnttab callback work. */
617 gnttab_cancel_free_callback(&info->callback);
618 spin_unlock_irqrestore(&blkif_io_lock, flags);
619
620 /* Flush gnttab callback work. Must be done with no locks held. */
621 flush_work_sync(&info->work);
622
623 del_gendisk(info->gd);
624
625 minor = info->gd->first_minor;
626 nr_minors = info->gd->minors;
627 xlbd_release_minors(minor, nr_minors);
628
629 blk_cleanup_queue(info->rq);
630 info->rq = NULL;
631
632 put_disk(info->gd);
633 info->gd = NULL;
634}
635
636static void kick_pending_request_queues(struct blkfront_info *info)
637{
638 if (!RING_FULL(&info->ring)) {
639 /* Re-enable calldowns. */
640 blk_start_queue(info->rq);
641 /* Kick things off immediately. */
642 do_blkif_request(info->rq);
643 }
644}
645
646static void blkif_restart_queue(struct work_struct *work)
647{
648 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
649
650 spin_lock_irq(&blkif_io_lock);
651 if (info->connected == BLKIF_STATE_CONNECTED)
652 kick_pending_request_queues(info);
653 spin_unlock_irq(&blkif_io_lock);
654}
655
656static void blkif_free(struct blkfront_info *info, int suspend)
657{
658 /* Prevent new requests being issued until we fix things up. */
659 spin_lock_irq(&blkif_io_lock);
660 info->connected = suspend ?
661 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
662 /* No more blkif_request(). */
663 if (info->rq)
664 blk_stop_queue(info->rq);
665 /* No more gnttab callback work. */
666 gnttab_cancel_free_callback(&info->callback);
667 spin_unlock_irq(&blkif_io_lock);
668
669 /* Flush gnttab callback work. Must be done with no locks held. */
670 flush_work_sync(&info->work);
671
672 /* Free resources associated with old device channel. */
673 if (info->ring_ref != GRANT_INVALID_REF) {
674 gnttab_end_foreign_access(info->ring_ref, 0,
675 (unsigned long)info->ring.sring);
676 info->ring_ref = GRANT_INVALID_REF;
677 info->ring.sring = NULL;
678 }
679 if (info->irq)
680 unbind_from_irqhandler(info->irq, info);
681 info->evtchn = info->irq = 0;
682
683}
684
685static void blkif_completion(struct blk_shadow *s)
686{
687 int i;
688 for (i = 0; i < s->req.nr_segments; i++)
689 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
690}
691
692static irqreturn_t blkif_interrupt(int irq, void *dev_id)
693{
694 struct request *req;
695 struct blkif_response *bret;
696 RING_IDX i, rp;
697 unsigned long flags;
698 struct blkfront_info *info = (struct blkfront_info *)dev_id;
699 int error;
700
701 spin_lock_irqsave(&blkif_io_lock, flags);
702
703 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
704 spin_unlock_irqrestore(&blkif_io_lock, flags);
705 return IRQ_HANDLED;
706 }
707
708 again:
709 rp = info->ring.sring->rsp_prod;
710 rmb(); /* Ensure we see queued responses up to 'rp'. */
711
712 for (i = info->ring.rsp_cons; i != rp; i++) {
713 unsigned long id;
714
715 bret = RING_GET_RESPONSE(&info->ring, i);
716 id = bret->id;
717 req = info->shadow[id].request;
718
719 blkif_completion(&info->shadow[id]);
720
721 add_id_to_freelist(info, id);
722
723 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
724 switch (bret->operation) {
725 case BLKIF_OP_FLUSH_DISKCACHE:
726 case BLKIF_OP_WRITE_BARRIER:
727 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
728 printk(KERN_WARNING "blkfront: %s: write %s op failed\n",
729 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
730 "barrier" : "flush disk cache",
731 info->gd->disk_name);
732 error = -EOPNOTSUPP;
733 }
734 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
735 info->shadow[id].req.nr_segments == 0)) {
736 printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n",
737 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
738 "barrier" : "flush disk cache",
739 info->gd->disk_name);
740 error = -EOPNOTSUPP;
741 }
742 if (unlikely(error)) {
743 if (error == -EOPNOTSUPP)
744 error = 0;
745 info->feature_flush = 0;
746 info->flush_op = 0;
747 xlvbd_flush(info);
748 }
749 /* fall through */
750 case BLKIF_OP_READ:
751 case BLKIF_OP_WRITE:
752 if (unlikely(bret->status != BLKIF_RSP_OKAY))
753 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
754 "request: %x\n", bret->status);
755
756 __blk_end_request_all(req, error);
757 break;
758 default:
759 BUG();
760 }
761 }
762
763 info->ring.rsp_cons = i;
764
765 if (i != info->ring.req_prod_pvt) {
766 int more_to_do;
767 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
768 if (more_to_do)
769 goto again;
770 } else
771 info->ring.sring->rsp_event = i + 1;
772
773 kick_pending_request_queues(info);
774
775 spin_unlock_irqrestore(&blkif_io_lock, flags);
776
777 return IRQ_HANDLED;
778}
779
780
781static int setup_blkring(struct xenbus_device *dev,
782 struct blkfront_info *info)
783{
784 struct blkif_sring *sring;
785 int err;
786
787 info->ring_ref = GRANT_INVALID_REF;
788
789 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
790 if (!sring) {
791 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
792 return -ENOMEM;
793 }
794 SHARED_RING_INIT(sring);
795 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
796
797 sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
798
799 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
800 if (err < 0) {
801 free_page((unsigned long)sring);
802 info->ring.sring = NULL;
803 goto fail;
804 }
805 info->ring_ref = err;
806
807 err = xenbus_alloc_evtchn(dev, &info->evtchn);
808 if (err)
809 goto fail;
810
811 err = bind_evtchn_to_irqhandler(info->evtchn,
812 blkif_interrupt,
813 IRQF_SAMPLE_RANDOM, "blkif", info);
814 if (err <= 0) {
815 xenbus_dev_fatal(dev, err,
816 "bind_evtchn_to_irqhandler failed");
817 goto fail;
818 }
819 info->irq = err;
820
821 return 0;
822fail:
823 blkif_free(info, 0);
824 return err;
825}
826
827
828/* Common code used when first setting up, and when resuming. */
829static int talk_to_blkback(struct xenbus_device *dev,
830 struct blkfront_info *info)
831{
832 const char *message = NULL;
833 struct xenbus_transaction xbt;
834 int err;
835
836 /* Create shared ring, alloc event channel. */
837 err = setup_blkring(dev, info);
838 if (err)
839 goto out;
840
841again:
842 err = xenbus_transaction_start(&xbt);
843 if (err) {
844 xenbus_dev_fatal(dev, err, "starting transaction");
845 goto destroy_blkring;
846 }
847
848 err = xenbus_printf(xbt, dev->nodename,
849 "ring-ref", "%u", info->ring_ref);
850 if (err) {
851 message = "writing ring-ref";
852 goto abort_transaction;
853 }
854 err = xenbus_printf(xbt, dev->nodename,
855 "event-channel", "%u", info->evtchn);
856 if (err) {
857 message = "writing event-channel";
858 goto abort_transaction;
859 }
860 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
861 XEN_IO_PROTO_ABI_NATIVE);
862 if (err) {
863 message = "writing protocol";
864 goto abort_transaction;
865 }
866
867 err = xenbus_transaction_end(xbt, 0);
868 if (err) {
869 if (err == -EAGAIN)
870 goto again;
871 xenbus_dev_fatal(dev, err, "completing transaction");
872 goto destroy_blkring;
873 }
874
875 xenbus_switch_state(dev, XenbusStateInitialised);
876
877 return 0;
878
879 abort_transaction:
880 xenbus_transaction_end(xbt, 1);
881 if (message)
882 xenbus_dev_fatal(dev, err, "%s", message);
883 destroy_blkring:
884 blkif_free(info, 0);
885 out:
886 return err;
887}
888
889/**
890 * Entry point to this code when a new device is created. Allocate the basic
891 * structures and the ring buffer for communication with the backend, and
892 * inform the backend of the appropriate details for those. Switch to
893 * Initialised state.
894 */
895static int blkfront_probe(struct xenbus_device *dev,
896 const struct xenbus_device_id *id)
897{
898 int err, vdevice, i;
899 struct blkfront_info *info;
900
901 /* FIXME: Use dynamic device id if this is not set. */
902 err = xenbus_scanf(XBT_NIL, dev->nodename,
903 "virtual-device", "%i", &vdevice);
904 if (err != 1) {
905 /* go looking in the extended area instead */
906 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
907 "%i", &vdevice);
908 if (err != 1) {
909 xenbus_dev_fatal(dev, err, "reading virtual-device");
910 return err;
911 }
912 }
913
914 if (xen_hvm_domain()) {
915 char *type;
916 int len;
917 /* no unplug has been done: do not hook devices != xen vbds */
918 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
919 int major;
920
921 if (!VDEV_IS_EXTENDED(vdevice))
922 major = BLKIF_MAJOR(vdevice);
923 else
924 major = XENVBD_MAJOR;
925
926 if (major != XENVBD_MAJOR) {
927 printk(KERN_INFO
928 "%s: HVM does not support vbd %d as xen block device\n",
929 __FUNCTION__, vdevice);
930 return -ENODEV;
931 }
932 }
933 /* do not create a PV cdrom device if we are an HVM guest */
934 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
935 if (IS_ERR(type))
936 return -ENODEV;
937 if (strncmp(type, "cdrom", 5) == 0) {
938 kfree(type);
939 return -ENODEV;
940 }
941 kfree(type);
942 }
943 info = kzalloc(sizeof(*info), GFP_KERNEL);
944 if (!info) {
945 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
946 return -ENOMEM;
947 }
948
949 mutex_init(&info->mutex);
950 info->xbdev = dev;
951 info->vdevice = vdevice;
952 info->connected = BLKIF_STATE_DISCONNECTED;
953 INIT_WORK(&info->work, blkif_restart_queue);
954
955 for (i = 0; i < BLK_RING_SIZE; i++)
956 info->shadow[i].req.id = i+1;
957 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
958
959 /* Front end dir is a number, which is used as the id. */
960 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
961 dev_set_drvdata(&dev->dev, info);
962
963 err = talk_to_blkback(dev, info);
964 if (err) {
965 kfree(info);
966 dev_set_drvdata(&dev->dev, NULL);
967 return err;
968 }
969
970 return 0;
971}
972
973
974static int blkif_recover(struct blkfront_info *info)
975{
976 int i;
977 struct blkif_request *req;
978 struct blk_shadow *copy;
979 int j;
980
981 /* Stage 1: Make a safe copy of the shadow state. */
982 copy = kmalloc(sizeof(info->shadow),
983 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
984 if (!copy)
985 return -ENOMEM;
986 memcpy(copy, info->shadow, sizeof(info->shadow));
987
988 /* Stage 2: Set up free list. */
989 memset(&info->shadow, 0, sizeof(info->shadow));
990 for (i = 0; i < BLK_RING_SIZE; i++)
991 info->shadow[i].req.id = i+1;
992 info->shadow_free = info->ring.req_prod_pvt;
993 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
994
995 /* Stage 3: Find pending requests and requeue them. */
996 for (i = 0; i < BLK_RING_SIZE; i++) {
997 /* Not in use? */
998 if (!copy[i].request)
999 continue;
1000
1001 /* Grab a request slot and copy shadow state into it. */
1002 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1003 *req = copy[i].req;
1004
1005 /* We get a new request id, and must reset the shadow state. */
1006 req->id = get_id_from_freelist(info);
1007 memcpy(&info->shadow[req->id], ©[i], sizeof(copy[i]));
1008
1009 /* Rewrite any grant references invalidated by susp/resume. */
1010 for (j = 0; j < req->nr_segments; j++)
1011 gnttab_grant_foreign_access_ref(
1012 req->u.rw.seg[j].gref,
1013 info->xbdev->otherend_id,
1014 pfn_to_mfn(info->shadow[req->id].frame[j]),
1015 rq_data_dir(info->shadow[req->id].request));
1016 info->shadow[req->id].req = *req;
1017
1018 info->ring.req_prod_pvt++;
1019 }
1020
1021 kfree(copy);
1022
1023 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1024
1025 spin_lock_irq(&blkif_io_lock);
1026
1027 /* Now safe for us to use the shared ring */
1028 info->connected = BLKIF_STATE_CONNECTED;
1029
1030 /* Send off requeued requests */
1031 flush_requests(info);
1032
1033 /* Kick any other new requests queued since we resumed */
1034 kick_pending_request_queues(info);
1035
1036 spin_unlock_irq(&blkif_io_lock);
1037
1038 return 0;
1039}
1040
1041/**
1042 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1043 * driver restart. We tear down our blkif structure and recreate it, but
1044 * leave the device-layer structures intact so that this is transparent to the
1045 * rest of the kernel.
1046 */
1047static int blkfront_resume(struct xenbus_device *dev)
1048{
1049 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1050 int err;
1051
1052 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1053
1054 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1055
1056 err = talk_to_blkback(dev, info);
1057 if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1058 err = blkif_recover(info);
1059
1060 return err;
1061}
1062
1063static void
1064blkfront_closing(struct blkfront_info *info)
1065{
1066 struct xenbus_device *xbdev = info->xbdev;
1067 struct block_device *bdev = NULL;
1068
1069 mutex_lock(&info->mutex);
1070
1071 if (xbdev->state == XenbusStateClosing) {
1072 mutex_unlock(&info->mutex);
1073 return;
1074 }
1075
1076 if (info->gd)
1077 bdev = bdget_disk(info->gd, 0);
1078
1079 mutex_unlock(&info->mutex);
1080
1081 if (!bdev) {
1082 xenbus_frontend_closed(xbdev);
1083 return;
1084 }
1085
1086 mutex_lock(&bdev->bd_mutex);
1087
1088 if (bdev->bd_openers) {
1089 xenbus_dev_error(xbdev, -EBUSY,
1090 "Device in use; refusing to close");
1091 xenbus_switch_state(xbdev, XenbusStateClosing);
1092 } else {
1093 xlvbd_release_gendisk(info);
1094 xenbus_frontend_closed(xbdev);
1095 }
1096
1097 mutex_unlock(&bdev->bd_mutex);
1098 bdput(bdev);
1099}
1100
1101/*
1102 * Invoked when the backend is finally 'ready' (and has told produced
1103 * the details about the physical device - #sectors, size, etc).
1104 */
1105static void blkfront_connect(struct blkfront_info *info)
1106{
1107 unsigned long long sectors;
1108 unsigned long sector_size;
1109 unsigned int binfo;
1110 int err;
1111 int barrier, flush;
1112
1113 switch (info->connected) {
1114 case BLKIF_STATE_CONNECTED:
1115 /*
1116 * Potentially, the back-end may be signalling
1117 * a capacity change; update the capacity.
1118 */
1119 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1120 "sectors", "%Lu", §ors);
1121 if (XENBUS_EXIST_ERR(err))
1122 return;
1123 printk(KERN_INFO "Setting capacity to %Lu\n",
1124 sectors);
1125 set_capacity(info->gd, sectors);
1126 revalidate_disk(info->gd);
1127
1128 /* fall through */
1129 case BLKIF_STATE_SUSPENDED:
1130 return;
1131
1132 default:
1133 break;
1134 }
1135
1136 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1137 __func__, info->xbdev->otherend);
1138
1139 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1140 "sectors", "%llu", §ors,
1141 "info", "%u", &binfo,
1142 "sector-size", "%lu", §or_size,
1143 NULL);
1144 if (err) {
1145 xenbus_dev_fatal(info->xbdev, err,
1146 "reading backend fields at %s",
1147 info->xbdev->otherend);
1148 return;
1149 }
1150
1151 info->feature_flush = 0;
1152 info->flush_op = 0;
1153
1154 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1155 "feature-barrier", "%d", &barrier,
1156 NULL);
1157
1158 /*
1159 * If there's no "feature-barrier" defined, then it means
1160 * we're dealing with a very old backend which writes
1161 * synchronously; nothing to do.
1162 *
1163 * If there are barriers, then we use flush.
1164 */
1165 if (!err && barrier) {
1166 info->feature_flush = REQ_FLUSH | REQ_FUA;
1167 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1168 }
1169 /*
1170 * And if there is "feature-flush-cache" use that above
1171 * barriers.
1172 */
1173 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1174 "feature-flush-cache", "%d", &flush,
1175 NULL);
1176
1177 if (!err && flush) {
1178 info->feature_flush = REQ_FLUSH;
1179 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1180 }
1181
1182 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1183 if (err) {
1184 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1185 info->xbdev->otherend);
1186 return;
1187 }
1188
1189 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1190
1191 /* Kick pending requests. */
1192 spin_lock_irq(&blkif_io_lock);
1193 info->connected = BLKIF_STATE_CONNECTED;
1194 kick_pending_request_queues(info);
1195 spin_unlock_irq(&blkif_io_lock);
1196
1197 add_disk(info->gd);
1198
1199 info->is_ready = 1;
1200}
1201
1202/**
1203 * Callback received when the backend's state changes.
1204 */
1205static void blkback_changed(struct xenbus_device *dev,
1206 enum xenbus_state backend_state)
1207{
1208 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1209
1210 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1211
1212 switch (backend_state) {
1213 case XenbusStateInitialising:
1214 case XenbusStateInitWait:
1215 case XenbusStateInitialised:
1216 case XenbusStateReconfiguring:
1217 case XenbusStateReconfigured:
1218 case XenbusStateUnknown:
1219 case XenbusStateClosed:
1220 break;
1221
1222 case XenbusStateConnected:
1223 blkfront_connect(info);
1224 break;
1225
1226 case XenbusStateClosing:
1227 blkfront_closing(info);
1228 break;
1229 }
1230}
1231
1232static int blkfront_remove(struct xenbus_device *xbdev)
1233{
1234 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1235 struct block_device *bdev = NULL;
1236 struct gendisk *disk;
1237
1238 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1239
1240 blkif_free(info, 0);
1241
1242 mutex_lock(&info->mutex);
1243
1244 disk = info->gd;
1245 if (disk)
1246 bdev = bdget_disk(disk, 0);
1247
1248 info->xbdev = NULL;
1249 mutex_unlock(&info->mutex);
1250
1251 if (!bdev) {
1252 kfree(info);
1253 return 0;
1254 }
1255
1256 /*
1257 * The xbdev was removed before we reached the Closed
1258 * state. See if it's safe to remove the disk. If the bdev
1259 * isn't closed yet, we let release take care of it.
1260 */
1261
1262 mutex_lock(&bdev->bd_mutex);
1263 info = disk->private_data;
1264
1265 dev_warn(disk_to_dev(disk),
1266 "%s was hot-unplugged, %d stale handles\n",
1267 xbdev->nodename, bdev->bd_openers);
1268
1269 if (info && !bdev->bd_openers) {
1270 xlvbd_release_gendisk(info);
1271 disk->private_data = NULL;
1272 kfree(info);
1273 }
1274
1275 mutex_unlock(&bdev->bd_mutex);
1276 bdput(bdev);
1277
1278 return 0;
1279}
1280
1281static int blkfront_is_ready(struct xenbus_device *dev)
1282{
1283 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1284
1285 return info->is_ready && info->xbdev;
1286}
1287
1288static int blkif_open(struct block_device *bdev, fmode_t mode)
1289{
1290 struct gendisk *disk = bdev->bd_disk;
1291 struct blkfront_info *info;
1292 int err = 0;
1293
1294 mutex_lock(&blkfront_mutex);
1295
1296 info = disk->private_data;
1297 if (!info) {
1298 /* xbdev gone */
1299 err = -ERESTARTSYS;
1300 goto out;
1301 }
1302
1303 mutex_lock(&info->mutex);
1304
1305 if (!info->gd)
1306 /* xbdev is closed */
1307 err = -ERESTARTSYS;
1308
1309 mutex_unlock(&info->mutex);
1310
1311out:
1312 mutex_unlock(&blkfront_mutex);
1313 return err;
1314}
1315
1316static int blkif_release(struct gendisk *disk, fmode_t mode)
1317{
1318 struct blkfront_info *info = disk->private_data;
1319 struct block_device *bdev;
1320 struct xenbus_device *xbdev;
1321
1322 mutex_lock(&blkfront_mutex);
1323
1324 bdev = bdget_disk(disk, 0);
1325 bdput(bdev);
1326
1327 if (bdev->bd_openers)
1328 goto out;
1329
1330 /*
1331 * Check if we have been instructed to close. We will have
1332 * deferred this request, because the bdev was still open.
1333 */
1334
1335 mutex_lock(&info->mutex);
1336 xbdev = info->xbdev;
1337
1338 if (xbdev && xbdev->state == XenbusStateClosing) {
1339 /* pending switch to state closed */
1340 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1341 xlvbd_release_gendisk(info);
1342 xenbus_frontend_closed(info->xbdev);
1343 }
1344
1345 mutex_unlock(&info->mutex);
1346
1347 if (!xbdev) {
1348 /* sudden device removal */
1349 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1350 xlvbd_release_gendisk(info);
1351 disk->private_data = NULL;
1352 kfree(info);
1353 }
1354
1355out:
1356 mutex_unlock(&blkfront_mutex);
1357 return 0;
1358}
1359
1360static const struct block_device_operations xlvbd_block_fops =
1361{
1362 .owner = THIS_MODULE,
1363 .open = blkif_open,
1364 .release = blkif_release,
1365 .getgeo = blkif_getgeo,
1366 .ioctl = blkif_ioctl,
1367};
1368
1369
1370static const struct xenbus_device_id blkfront_ids[] = {
1371 { "vbd" },
1372 { "" }
1373};
1374
1375static struct xenbus_driver blkfront = {
1376 .name = "vbd",
1377 .owner = THIS_MODULE,
1378 .ids = blkfront_ids,
1379 .probe = blkfront_probe,
1380 .remove = blkfront_remove,
1381 .resume = blkfront_resume,
1382 .otherend_changed = blkback_changed,
1383 .is_ready = blkfront_is_ready,
1384};
1385
1386static int __init xlblk_init(void)
1387{
1388 if (!xen_domain())
1389 return -ENODEV;
1390
1391 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1392 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1393 XENVBD_MAJOR, DEV_NAME);
1394 return -ENODEV;
1395 }
1396
1397 return xenbus_register_frontend(&blkfront);
1398}
1399module_init(xlblk_init);
1400
1401
1402static void __exit xlblk_exit(void)
1403{
1404 return xenbus_unregister_driver(&blkfront);
1405}
1406module_exit(xlblk_exit);
1407
1408MODULE_DESCRIPTION("Xen virtual block device frontend");
1409MODULE_LICENSE("GPL");
1410MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1411MODULE_ALIAS("xen:vbd");
1412MODULE_ALIAS("xenblk");
1/*
2 * blkfront.c
3 *
4 * XenLinux virtual block device driver.
5 *
6 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8 * Copyright (c) 2004, Christian Limpach
9 * Copyright (c) 2004, Andrew Warfield
10 * Copyright (c) 2005, Christopher Clark
11 * Copyright (c) 2005, XenSource Ltd
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License version 2
15 * as published by the Free Software Foundation; or, when distributed
16 * separately from the Linux kernel or incorporated into other
17 * software packages, subject to the following license:
18 *
19 * Permission is hereby granted, free of charge, to any person obtaining a copy
20 * of this source file (the "Software"), to deal in the Software without
21 * restriction, including without limitation the rights to use, copy, modify,
22 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23 * and to permit persons to whom the Software is furnished to do so, subject to
24 * the following conditions:
25 *
26 * The above copyright notice and this permission notice shall be included in
27 * all copies or substantial portions of the Software.
28 *
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
35 * IN THE SOFTWARE.
36 */
37
38#include <linux/interrupt.h>
39#include <linux/blkdev.h>
40#include <linux/blk-mq.h>
41#include <linux/hdreg.h>
42#include <linux/cdrom.h>
43#include <linux/module.h>
44#include <linux/slab.h>
45#include <linux/major.h>
46#include <linux/mutex.h>
47#include <linux/scatterlist.h>
48#include <linux/bitmap.h>
49#include <linux/list.h>
50#include <linux/workqueue.h>
51#include <linux/sched/mm.h>
52
53#include <xen/xen.h>
54#include <xen/xenbus.h>
55#include <xen/grant_table.h>
56#include <xen/events.h>
57#include <xen/page.h>
58#include <xen/platform_pci.h>
59
60#include <xen/interface/grant_table.h>
61#include <xen/interface/io/blkif.h>
62#include <xen/interface/io/protocols.h>
63
64#include <asm/xen/hypervisor.h>
65
66/*
67 * The minimal size of segment supported by the block framework is PAGE_SIZE.
68 * When Linux is using a different page size than Xen, it may not be possible
69 * to put all the data in a single segment.
70 * This can happen when the backend doesn't support indirect descriptor and
71 * therefore the maximum amount of data that a request can carry is
72 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
73 *
74 * Note that we only support one extra request. So the Linux page size
75 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
76 * 88KB.
77 */
78#define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
79
80enum blkif_state {
81 BLKIF_STATE_DISCONNECTED,
82 BLKIF_STATE_CONNECTED,
83 BLKIF_STATE_SUSPENDED,
84 BLKIF_STATE_ERROR,
85};
86
87struct grant {
88 grant_ref_t gref;
89 struct page *page;
90 struct list_head node;
91};
92
93enum blk_req_status {
94 REQ_PROCESSING,
95 REQ_WAITING,
96 REQ_DONE,
97 REQ_ERROR,
98 REQ_EOPNOTSUPP,
99};
100
101struct blk_shadow {
102 struct blkif_request req;
103 struct request *request;
104 struct grant **grants_used;
105 struct grant **indirect_grants;
106 struct scatterlist *sg;
107 unsigned int num_sg;
108 enum blk_req_status status;
109
110 #define NO_ASSOCIATED_ID ~0UL
111 /*
112 * Id of the sibling if we ever need 2 requests when handling a
113 * block I/O request
114 */
115 unsigned long associated_id;
116};
117
118struct blkif_req {
119 blk_status_t error;
120};
121
122static inline struct blkif_req *blkif_req(struct request *rq)
123{
124 return blk_mq_rq_to_pdu(rq);
125}
126
127static DEFINE_MUTEX(blkfront_mutex);
128static const struct block_device_operations xlvbd_block_fops;
129static struct delayed_work blkfront_work;
130static LIST_HEAD(info_list);
131
132/*
133 * Maximum number of segments in indirect requests, the actual value used by
134 * the frontend driver is the minimum of this value and the value provided
135 * by the backend driver.
136 */
137
138static unsigned int xen_blkif_max_segments = 32;
139module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
140MODULE_PARM_DESC(max_indirect_segments,
141 "Maximum amount of segments in indirect requests (default is 32)");
142
143static unsigned int xen_blkif_max_queues = 4;
144module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
145MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
146
147/*
148 * Maximum order of pages to be used for the shared ring between front and
149 * backend, 4KB page granularity is used.
150 */
151static unsigned int xen_blkif_max_ring_order;
152module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
153MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
154
155static bool __read_mostly xen_blkif_trusted = true;
156module_param_named(trusted, xen_blkif_trusted, bool, 0644);
157MODULE_PARM_DESC(trusted, "Is the backend trusted");
158
159#define BLK_RING_SIZE(info) \
160 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
161
162/*
163 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
164 * characters are enough. Define to 20 to keep consistent with backend.
165 */
166#define RINGREF_NAME_LEN (20)
167/*
168 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
169 */
170#define QUEUE_NAME_LEN (17)
171
172/*
173 * Per-ring info.
174 * Every blkfront device can associate with one or more blkfront_ring_info,
175 * depending on how many hardware queues/rings to be used.
176 */
177struct blkfront_ring_info {
178 /* Lock to protect data in every ring buffer. */
179 spinlock_t ring_lock;
180 struct blkif_front_ring ring;
181 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
182 unsigned int evtchn, irq;
183 struct work_struct work;
184 struct gnttab_free_callback callback;
185 struct list_head indirect_pages;
186 struct list_head grants;
187 unsigned int persistent_gnts_c;
188 unsigned long shadow_free;
189 struct blkfront_info *dev_info;
190 struct blk_shadow shadow[];
191};
192
193/*
194 * We have one of these per vbd, whether ide, scsi or 'other'. They
195 * hang in private_data off the gendisk structure. We may end up
196 * putting all kinds of interesting stuff here :-)
197 */
198struct blkfront_info
199{
200 struct mutex mutex;
201 struct xenbus_device *xbdev;
202 struct gendisk *gd;
203 u16 sector_size;
204 unsigned int physical_sector_size;
205 unsigned long vdisk_info;
206 int vdevice;
207 blkif_vdev_t handle;
208 enum blkif_state connected;
209 /* Number of pages per ring buffer. */
210 unsigned int nr_ring_pages;
211 struct request_queue *rq;
212 unsigned int feature_flush:1;
213 unsigned int feature_fua:1;
214 unsigned int feature_discard:1;
215 unsigned int feature_secdiscard:1;
216 /* Connect-time cached feature_persistent parameter */
217 unsigned int feature_persistent_parm:1;
218 /* Persistent grants feature negotiation result */
219 unsigned int feature_persistent:1;
220 unsigned int bounce:1;
221 unsigned int discard_granularity;
222 unsigned int discard_alignment;
223 /* Number of 4KB segments handled */
224 unsigned int max_indirect_segments;
225 int is_ready;
226 struct blk_mq_tag_set tag_set;
227 struct blkfront_ring_info *rinfo;
228 unsigned int nr_rings;
229 unsigned int rinfo_size;
230 /* Save uncomplete reqs and bios for migration. */
231 struct list_head requests;
232 struct bio_list bio_list;
233 struct list_head info_list;
234};
235
236static unsigned int nr_minors;
237static unsigned long *minors;
238static DEFINE_SPINLOCK(minor_lock);
239
240#define PARTS_PER_DISK 16
241#define PARTS_PER_EXT_DISK 256
242
243#define BLKIF_MAJOR(dev) ((dev)>>8)
244#define BLKIF_MINOR(dev) ((dev) & 0xff)
245
246#define EXT_SHIFT 28
247#define EXTENDED (1<<EXT_SHIFT)
248#define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
249#define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
250#define EMULATED_HD_DISK_MINOR_OFFSET (0)
251#define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
252#define EMULATED_SD_DISK_MINOR_OFFSET (0)
253#define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
254
255#define DEV_NAME "xvd" /* name in /dev */
256
257/*
258 * Grants are always the same size as a Xen page (i.e 4KB).
259 * A physical segment is always the same size as a Linux page.
260 * Number of grants per physical segment
261 */
262#define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
263
264#define GRANTS_PER_INDIRECT_FRAME \
265 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
266
267#define INDIRECT_GREFS(_grants) \
268 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
269
270static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
271static void blkfront_gather_backend_features(struct blkfront_info *info);
272static int negotiate_mq(struct blkfront_info *info);
273
274#define for_each_rinfo(info, ptr, idx) \
275 for ((ptr) = (info)->rinfo, (idx) = 0; \
276 (idx) < (info)->nr_rings; \
277 (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
278
279static inline struct blkfront_ring_info *
280get_rinfo(const struct blkfront_info *info, unsigned int i)
281{
282 BUG_ON(i >= info->nr_rings);
283 return (void *)info->rinfo + i * info->rinfo_size;
284}
285
286static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
287{
288 unsigned long free = rinfo->shadow_free;
289
290 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
291 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
292 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
293 return free;
294}
295
296static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
297 unsigned long id)
298{
299 if (rinfo->shadow[id].req.u.rw.id != id)
300 return -EINVAL;
301 if (rinfo->shadow[id].request == NULL)
302 return -EINVAL;
303 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
304 rinfo->shadow[id].request = NULL;
305 rinfo->shadow_free = id;
306 return 0;
307}
308
309static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
310{
311 struct blkfront_info *info = rinfo->dev_info;
312 struct page *granted_page;
313 struct grant *gnt_list_entry, *n;
314 int i = 0;
315
316 while (i < num) {
317 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
318 if (!gnt_list_entry)
319 goto out_of_memory;
320
321 if (info->bounce) {
322 granted_page = alloc_page(GFP_NOIO | __GFP_ZERO);
323 if (!granted_page) {
324 kfree(gnt_list_entry);
325 goto out_of_memory;
326 }
327 gnt_list_entry->page = granted_page;
328 }
329
330 gnt_list_entry->gref = INVALID_GRANT_REF;
331 list_add(&gnt_list_entry->node, &rinfo->grants);
332 i++;
333 }
334
335 return 0;
336
337out_of_memory:
338 list_for_each_entry_safe(gnt_list_entry, n,
339 &rinfo->grants, node) {
340 list_del(&gnt_list_entry->node);
341 if (info->bounce)
342 __free_page(gnt_list_entry->page);
343 kfree(gnt_list_entry);
344 i--;
345 }
346 BUG_ON(i != 0);
347 return -ENOMEM;
348}
349
350static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
351{
352 struct grant *gnt_list_entry;
353
354 BUG_ON(list_empty(&rinfo->grants));
355 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
356 node);
357 list_del(&gnt_list_entry->node);
358
359 if (gnt_list_entry->gref != INVALID_GRANT_REF)
360 rinfo->persistent_gnts_c--;
361
362 return gnt_list_entry;
363}
364
365static inline void grant_foreign_access(const struct grant *gnt_list_entry,
366 const struct blkfront_info *info)
367{
368 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
369 info->xbdev->otherend_id,
370 gnt_list_entry->page,
371 0);
372}
373
374static struct grant *get_grant(grant_ref_t *gref_head,
375 unsigned long gfn,
376 struct blkfront_ring_info *rinfo)
377{
378 struct grant *gnt_list_entry = get_free_grant(rinfo);
379 struct blkfront_info *info = rinfo->dev_info;
380
381 if (gnt_list_entry->gref != INVALID_GRANT_REF)
382 return gnt_list_entry;
383
384 /* Assign a gref to this page */
385 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
386 BUG_ON(gnt_list_entry->gref == -ENOSPC);
387 if (info->bounce)
388 grant_foreign_access(gnt_list_entry, info);
389 else {
390 /* Grant access to the GFN passed by the caller */
391 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
392 info->xbdev->otherend_id,
393 gfn, 0);
394 }
395
396 return gnt_list_entry;
397}
398
399static struct grant *get_indirect_grant(grant_ref_t *gref_head,
400 struct blkfront_ring_info *rinfo)
401{
402 struct grant *gnt_list_entry = get_free_grant(rinfo);
403 struct blkfront_info *info = rinfo->dev_info;
404
405 if (gnt_list_entry->gref != INVALID_GRANT_REF)
406 return gnt_list_entry;
407
408 /* Assign a gref to this page */
409 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
410 BUG_ON(gnt_list_entry->gref == -ENOSPC);
411 if (!info->bounce) {
412 struct page *indirect_page;
413
414 /* Fetch a pre-allocated page to use for indirect grefs */
415 BUG_ON(list_empty(&rinfo->indirect_pages));
416 indirect_page = list_first_entry(&rinfo->indirect_pages,
417 struct page, lru);
418 list_del(&indirect_page->lru);
419 gnt_list_entry->page = indirect_page;
420 }
421 grant_foreign_access(gnt_list_entry, info);
422
423 return gnt_list_entry;
424}
425
426static const char *op_name(int op)
427{
428 static const char *const names[] = {
429 [BLKIF_OP_READ] = "read",
430 [BLKIF_OP_WRITE] = "write",
431 [BLKIF_OP_WRITE_BARRIER] = "barrier",
432 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
433 [BLKIF_OP_DISCARD] = "discard" };
434
435 if (op < 0 || op >= ARRAY_SIZE(names))
436 return "unknown";
437
438 if (!names[op])
439 return "reserved";
440
441 return names[op];
442}
443static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
444{
445 unsigned int end = minor + nr;
446 int rc;
447
448 if (end > nr_minors) {
449 unsigned long *bitmap, *old;
450
451 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
452 GFP_KERNEL);
453 if (bitmap == NULL)
454 return -ENOMEM;
455
456 spin_lock(&minor_lock);
457 if (end > nr_minors) {
458 old = minors;
459 memcpy(bitmap, minors,
460 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
461 minors = bitmap;
462 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
463 } else
464 old = bitmap;
465 spin_unlock(&minor_lock);
466 kfree(old);
467 }
468
469 spin_lock(&minor_lock);
470 if (find_next_bit(minors, end, minor) >= end) {
471 bitmap_set(minors, minor, nr);
472 rc = 0;
473 } else
474 rc = -EBUSY;
475 spin_unlock(&minor_lock);
476
477 return rc;
478}
479
480static void xlbd_release_minors(unsigned int minor, unsigned int nr)
481{
482 unsigned int end = minor + nr;
483
484 BUG_ON(end > nr_minors);
485 spin_lock(&minor_lock);
486 bitmap_clear(minors, minor, nr);
487 spin_unlock(&minor_lock);
488}
489
490static void blkif_restart_queue_callback(void *arg)
491{
492 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
493 schedule_work(&rinfo->work);
494}
495
496static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
497{
498 /* We don't have real geometry info, but let's at least return
499 values consistent with the size of the device */
500 sector_t nsect = get_capacity(bd->bd_disk);
501 sector_t cylinders = nsect;
502
503 hg->heads = 0xff;
504 hg->sectors = 0x3f;
505 sector_div(cylinders, hg->heads * hg->sectors);
506 hg->cylinders = cylinders;
507 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
508 hg->cylinders = 0xffff;
509 return 0;
510}
511
512static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
513 unsigned command, unsigned long argument)
514{
515 struct blkfront_info *info = bdev->bd_disk->private_data;
516 int i;
517
518 switch (command) {
519 case CDROMMULTISESSION:
520 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
521 if (put_user(0, (char __user *)(argument + i)))
522 return -EFAULT;
523 return 0;
524 case CDROM_GET_CAPABILITY:
525 if (!(info->vdisk_info & VDISK_CDROM))
526 return -EINVAL;
527 return 0;
528 default:
529 return -EINVAL;
530 }
531}
532
533static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
534 struct request *req,
535 struct blkif_request **ring_req)
536{
537 unsigned long id;
538
539 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
540 rinfo->ring.req_prod_pvt++;
541
542 id = get_id_from_freelist(rinfo);
543 rinfo->shadow[id].request = req;
544 rinfo->shadow[id].status = REQ_PROCESSING;
545 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
546
547 rinfo->shadow[id].req.u.rw.id = id;
548
549 return id;
550}
551
552static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
553{
554 struct blkfront_info *info = rinfo->dev_info;
555 struct blkif_request *ring_req, *final_ring_req;
556 unsigned long id;
557
558 /* Fill out a communications ring structure. */
559 id = blkif_ring_get_request(rinfo, req, &final_ring_req);
560 ring_req = &rinfo->shadow[id].req;
561
562 ring_req->operation = BLKIF_OP_DISCARD;
563 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
564 ring_req->u.discard.id = id;
565 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
566 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
567 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
568 else
569 ring_req->u.discard.flag = 0;
570
571 /* Copy the request to the ring page. */
572 *final_ring_req = *ring_req;
573 rinfo->shadow[id].status = REQ_WAITING;
574
575 return 0;
576}
577
578struct setup_rw_req {
579 unsigned int grant_idx;
580 struct blkif_request_segment *segments;
581 struct blkfront_ring_info *rinfo;
582 struct blkif_request *ring_req;
583 grant_ref_t gref_head;
584 unsigned int id;
585 /* Only used when persistent grant is used and it's a write request */
586 bool need_copy;
587 unsigned int bvec_off;
588 char *bvec_data;
589
590 bool require_extra_req;
591 struct blkif_request *extra_ring_req;
592};
593
594static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
595 unsigned int len, void *data)
596{
597 struct setup_rw_req *setup = data;
598 int n, ref;
599 struct grant *gnt_list_entry;
600 unsigned int fsect, lsect;
601 /* Convenient aliases */
602 unsigned int grant_idx = setup->grant_idx;
603 struct blkif_request *ring_req = setup->ring_req;
604 struct blkfront_ring_info *rinfo = setup->rinfo;
605 /*
606 * We always use the shadow of the first request to store the list
607 * of grant associated to the block I/O request. This made the
608 * completion more easy to handle even if the block I/O request is
609 * split.
610 */
611 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
612
613 if (unlikely(setup->require_extra_req &&
614 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
615 /*
616 * We are using the second request, setup grant_idx
617 * to be the index of the segment array.
618 */
619 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
620 ring_req = setup->extra_ring_req;
621 }
622
623 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
624 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
625 if (setup->segments)
626 kunmap_atomic(setup->segments);
627
628 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
629 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
630 shadow->indirect_grants[n] = gnt_list_entry;
631 setup->segments = kmap_atomic(gnt_list_entry->page);
632 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
633 }
634
635 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
636 ref = gnt_list_entry->gref;
637 /*
638 * All the grants are stored in the shadow of the first
639 * request. Therefore we have to use the global index.
640 */
641 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
642
643 if (setup->need_copy) {
644 void *shared_data;
645
646 shared_data = kmap_atomic(gnt_list_entry->page);
647 /*
648 * this does not wipe data stored outside the
649 * range sg->offset..sg->offset+sg->length.
650 * Therefore, blkback *could* see data from
651 * previous requests. This is OK as long as
652 * persistent grants are shared with just one
653 * domain. It may need refactoring if this
654 * changes
655 */
656 memcpy(shared_data + offset,
657 setup->bvec_data + setup->bvec_off,
658 len);
659
660 kunmap_atomic(shared_data);
661 setup->bvec_off += len;
662 }
663
664 fsect = offset >> 9;
665 lsect = fsect + (len >> 9) - 1;
666 if (ring_req->operation != BLKIF_OP_INDIRECT) {
667 ring_req->u.rw.seg[grant_idx] =
668 (struct blkif_request_segment) {
669 .gref = ref,
670 .first_sect = fsect,
671 .last_sect = lsect };
672 } else {
673 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
674 (struct blkif_request_segment) {
675 .gref = ref,
676 .first_sect = fsect,
677 .last_sect = lsect };
678 }
679
680 (setup->grant_idx)++;
681}
682
683static void blkif_setup_extra_req(struct blkif_request *first,
684 struct blkif_request *second)
685{
686 uint16_t nr_segments = first->u.rw.nr_segments;
687
688 /*
689 * The second request is only present when the first request uses
690 * all its segments. It's always the continuity of the first one.
691 */
692 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
693
694 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
695 second->u.rw.sector_number = first->u.rw.sector_number +
696 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
697
698 second->u.rw.handle = first->u.rw.handle;
699 second->operation = first->operation;
700}
701
702static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
703{
704 struct blkfront_info *info = rinfo->dev_info;
705 struct blkif_request *ring_req, *extra_ring_req = NULL;
706 struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
707 unsigned long id, extra_id = NO_ASSOCIATED_ID;
708 bool require_extra_req = false;
709 int i;
710 struct setup_rw_req setup = {
711 .grant_idx = 0,
712 .segments = NULL,
713 .rinfo = rinfo,
714 .need_copy = rq_data_dir(req) && info->bounce,
715 };
716
717 /*
718 * Used to store if we are able to queue the request by just using
719 * existing persistent grants, or if we have to get new grants,
720 * as there are not sufficiently many free.
721 */
722 bool new_persistent_gnts = false;
723 struct scatterlist *sg;
724 int num_sg, max_grefs, num_grant;
725
726 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
727 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
728 /*
729 * If we are using indirect segments we need to account
730 * for the indirect grefs used in the request.
731 */
732 max_grefs += INDIRECT_GREFS(max_grefs);
733
734 /* Check if we have enough persistent grants to allocate a requests */
735 if (rinfo->persistent_gnts_c < max_grefs) {
736 new_persistent_gnts = true;
737
738 if (gnttab_alloc_grant_references(
739 max_grefs - rinfo->persistent_gnts_c,
740 &setup.gref_head) < 0) {
741 gnttab_request_free_callback(
742 &rinfo->callback,
743 blkif_restart_queue_callback,
744 rinfo,
745 max_grefs - rinfo->persistent_gnts_c);
746 return 1;
747 }
748 }
749
750 /* Fill out a communications ring structure. */
751 id = blkif_ring_get_request(rinfo, req, &final_ring_req);
752 ring_req = &rinfo->shadow[id].req;
753
754 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
755 num_grant = 0;
756 /* Calculate the number of grant used */
757 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
758 num_grant += gnttab_count_grant(sg->offset, sg->length);
759
760 require_extra_req = info->max_indirect_segments == 0 &&
761 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
762 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
763
764 rinfo->shadow[id].num_sg = num_sg;
765 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
766 likely(!require_extra_req)) {
767 /*
768 * The indirect operation can only be a BLKIF_OP_READ or
769 * BLKIF_OP_WRITE
770 */
771 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
772 ring_req->operation = BLKIF_OP_INDIRECT;
773 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
774 BLKIF_OP_WRITE : BLKIF_OP_READ;
775 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
776 ring_req->u.indirect.handle = info->handle;
777 ring_req->u.indirect.nr_segments = num_grant;
778 } else {
779 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
780 ring_req->u.rw.handle = info->handle;
781 ring_req->operation = rq_data_dir(req) ?
782 BLKIF_OP_WRITE : BLKIF_OP_READ;
783 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
784 /*
785 * Ideally we can do an unordered flush-to-disk.
786 * In case the backend onlysupports barriers, use that.
787 * A barrier request a superset of FUA, so we can
788 * implement it the same way. (It's also a FLUSH+FUA,
789 * since it is guaranteed ordered WRT previous writes.)
790 */
791 if (info->feature_flush && info->feature_fua)
792 ring_req->operation =
793 BLKIF_OP_WRITE_BARRIER;
794 else if (info->feature_flush)
795 ring_req->operation =
796 BLKIF_OP_FLUSH_DISKCACHE;
797 else
798 ring_req->operation = 0;
799 }
800 ring_req->u.rw.nr_segments = num_grant;
801 if (unlikely(require_extra_req)) {
802 extra_id = blkif_ring_get_request(rinfo, req,
803 &final_extra_ring_req);
804 extra_ring_req = &rinfo->shadow[extra_id].req;
805
806 /*
807 * Only the first request contains the scatter-gather
808 * list.
809 */
810 rinfo->shadow[extra_id].num_sg = 0;
811
812 blkif_setup_extra_req(ring_req, extra_ring_req);
813
814 /* Link the 2 requests together */
815 rinfo->shadow[extra_id].associated_id = id;
816 rinfo->shadow[id].associated_id = extra_id;
817 }
818 }
819
820 setup.ring_req = ring_req;
821 setup.id = id;
822
823 setup.require_extra_req = require_extra_req;
824 if (unlikely(require_extra_req))
825 setup.extra_ring_req = extra_ring_req;
826
827 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
828 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
829
830 if (setup.need_copy) {
831 setup.bvec_off = sg->offset;
832 setup.bvec_data = kmap_atomic(sg_page(sg));
833 }
834
835 gnttab_foreach_grant_in_range(sg_page(sg),
836 sg->offset,
837 sg->length,
838 blkif_setup_rw_req_grant,
839 &setup);
840
841 if (setup.need_copy)
842 kunmap_atomic(setup.bvec_data);
843 }
844 if (setup.segments)
845 kunmap_atomic(setup.segments);
846
847 /* Copy request(s) to the ring page. */
848 *final_ring_req = *ring_req;
849 rinfo->shadow[id].status = REQ_WAITING;
850 if (unlikely(require_extra_req)) {
851 *final_extra_ring_req = *extra_ring_req;
852 rinfo->shadow[extra_id].status = REQ_WAITING;
853 }
854
855 if (new_persistent_gnts)
856 gnttab_free_grant_references(setup.gref_head);
857
858 return 0;
859}
860
861/*
862 * Generate a Xen blkfront IO request from a blk layer request. Reads
863 * and writes are handled as expected.
864 *
865 * @req: a request struct
866 */
867static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
868{
869 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
870 return 1;
871
872 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
873 req_op(req) == REQ_OP_SECURE_ERASE))
874 return blkif_queue_discard_req(req, rinfo);
875 else
876 return blkif_queue_rw_req(req, rinfo);
877}
878
879static inline void flush_requests(struct blkfront_ring_info *rinfo)
880{
881 int notify;
882
883 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
884
885 if (notify)
886 notify_remote_via_irq(rinfo->irq);
887}
888
889static inline bool blkif_request_flush_invalid(struct request *req,
890 struct blkfront_info *info)
891{
892 return (blk_rq_is_passthrough(req) ||
893 ((req_op(req) == REQ_OP_FLUSH) &&
894 !info->feature_flush) ||
895 ((req->cmd_flags & REQ_FUA) &&
896 !info->feature_fua));
897}
898
899static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
900 const struct blk_mq_queue_data *qd)
901{
902 unsigned long flags;
903 int qid = hctx->queue_num;
904 struct blkfront_info *info = hctx->queue->queuedata;
905 struct blkfront_ring_info *rinfo = NULL;
906
907 rinfo = get_rinfo(info, qid);
908 blk_mq_start_request(qd->rq);
909 spin_lock_irqsave(&rinfo->ring_lock, flags);
910 if (RING_FULL(&rinfo->ring))
911 goto out_busy;
912
913 if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
914 goto out_err;
915
916 if (blkif_queue_request(qd->rq, rinfo))
917 goto out_busy;
918
919 flush_requests(rinfo);
920 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
921 return BLK_STS_OK;
922
923out_err:
924 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
925 return BLK_STS_IOERR;
926
927out_busy:
928 blk_mq_stop_hw_queue(hctx);
929 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
930 return BLK_STS_DEV_RESOURCE;
931}
932
933static void blkif_complete_rq(struct request *rq)
934{
935 blk_mq_end_request(rq, blkif_req(rq)->error);
936}
937
938static const struct blk_mq_ops blkfront_mq_ops = {
939 .queue_rq = blkif_queue_rq,
940 .complete = blkif_complete_rq,
941};
942
943static void blkif_set_queue_limits(struct blkfront_info *info)
944{
945 struct request_queue *rq = info->rq;
946 struct gendisk *gd = info->gd;
947 unsigned int segments = info->max_indirect_segments ? :
948 BLKIF_MAX_SEGMENTS_PER_REQUEST;
949
950 blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
951
952 if (info->feature_discard) {
953 blk_queue_max_discard_sectors(rq, get_capacity(gd));
954 rq->limits.discard_granularity = info->discard_granularity ?:
955 info->physical_sector_size;
956 rq->limits.discard_alignment = info->discard_alignment;
957 if (info->feature_secdiscard)
958 blk_queue_max_secure_erase_sectors(rq,
959 get_capacity(gd));
960 }
961
962 /* Hard sector size and max sectors impersonate the equiv. hardware. */
963 blk_queue_logical_block_size(rq, info->sector_size);
964 blk_queue_physical_block_size(rq, info->physical_sector_size);
965 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
966
967 /* Each segment in a request is up to an aligned page in size. */
968 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
969 blk_queue_max_segment_size(rq, PAGE_SIZE);
970
971 /* Ensure a merged request will fit in a single I/O ring slot. */
972 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
973
974 /* Make sure buffer addresses are sector-aligned. */
975 blk_queue_dma_alignment(rq, 511);
976}
977
978static const char *flush_info(struct blkfront_info *info)
979{
980 if (info->feature_flush && info->feature_fua)
981 return "barrier: enabled;";
982 else if (info->feature_flush)
983 return "flush diskcache: enabled;";
984 else
985 return "barrier or flush: disabled;";
986}
987
988static void xlvbd_flush(struct blkfront_info *info)
989{
990 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
991 info->feature_fua ? true : false);
992 pr_info("blkfront: %s: %s %s %s %s %s %s %s\n",
993 info->gd->disk_name, flush_info(info),
994 "persistent grants:", info->feature_persistent ?
995 "enabled;" : "disabled;", "indirect descriptors:",
996 info->max_indirect_segments ? "enabled;" : "disabled;",
997 "bounce buffer:", info->bounce ? "enabled" : "disabled;");
998}
999
1000static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1001{
1002 int major;
1003 major = BLKIF_MAJOR(vdevice);
1004 *minor = BLKIF_MINOR(vdevice);
1005 switch (major) {
1006 case XEN_IDE0_MAJOR:
1007 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1008 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1009 EMULATED_HD_DISK_MINOR_OFFSET;
1010 break;
1011 case XEN_IDE1_MAJOR:
1012 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1013 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1014 EMULATED_HD_DISK_MINOR_OFFSET;
1015 break;
1016 case XEN_SCSI_DISK0_MAJOR:
1017 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1018 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1019 break;
1020 case XEN_SCSI_DISK1_MAJOR:
1021 case XEN_SCSI_DISK2_MAJOR:
1022 case XEN_SCSI_DISK3_MAJOR:
1023 case XEN_SCSI_DISK4_MAJOR:
1024 case XEN_SCSI_DISK5_MAJOR:
1025 case XEN_SCSI_DISK6_MAJOR:
1026 case XEN_SCSI_DISK7_MAJOR:
1027 *offset = (*minor / PARTS_PER_DISK) +
1028 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1029 EMULATED_SD_DISK_NAME_OFFSET;
1030 *minor = *minor +
1031 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1032 EMULATED_SD_DISK_MINOR_OFFSET;
1033 break;
1034 case XEN_SCSI_DISK8_MAJOR:
1035 case XEN_SCSI_DISK9_MAJOR:
1036 case XEN_SCSI_DISK10_MAJOR:
1037 case XEN_SCSI_DISK11_MAJOR:
1038 case XEN_SCSI_DISK12_MAJOR:
1039 case XEN_SCSI_DISK13_MAJOR:
1040 case XEN_SCSI_DISK14_MAJOR:
1041 case XEN_SCSI_DISK15_MAJOR:
1042 *offset = (*minor / PARTS_PER_DISK) +
1043 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1044 EMULATED_SD_DISK_NAME_OFFSET;
1045 *minor = *minor +
1046 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1047 EMULATED_SD_DISK_MINOR_OFFSET;
1048 break;
1049 case XENVBD_MAJOR:
1050 *offset = *minor / PARTS_PER_DISK;
1051 break;
1052 default:
1053 printk(KERN_WARNING "blkfront: your disk configuration is "
1054 "incorrect, please use an xvd device instead\n");
1055 return -ENODEV;
1056 }
1057 return 0;
1058}
1059
1060static char *encode_disk_name(char *ptr, unsigned int n)
1061{
1062 if (n >= 26)
1063 ptr = encode_disk_name(ptr, n / 26 - 1);
1064 *ptr = 'a' + n % 26;
1065 return ptr + 1;
1066}
1067
1068static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1069 struct blkfront_info *info, u16 sector_size,
1070 unsigned int physical_sector_size)
1071{
1072 struct gendisk *gd;
1073 int nr_minors = 1;
1074 int err;
1075 unsigned int offset;
1076 int minor;
1077 int nr_parts;
1078 char *ptr;
1079
1080 BUG_ON(info->gd != NULL);
1081 BUG_ON(info->rq != NULL);
1082
1083 if ((info->vdevice>>EXT_SHIFT) > 1) {
1084 /* this is above the extended range; something is wrong */
1085 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1086 return -ENODEV;
1087 }
1088
1089 if (!VDEV_IS_EXTENDED(info->vdevice)) {
1090 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1091 if (err)
1092 return err;
1093 nr_parts = PARTS_PER_DISK;
1094 } else {
1095 minor = BLKIF_MINOR_EXT(info->vdevice);
1096 nr_parts = PARTS_PER_EXT_DISK;
1097 offset = minor / nr_parts;
1098 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1099 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1100 "emulated IDE disks,\n\t choose an xvd device name"
1101 "from xvde on\n", info->vdevice);
1102 }
1103 if (minor >> MINORBITS) {
1104 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1105 info->vdevice, minor);
1106 return -ENODEV;
1107 }
1108
1109 if ((minor % nr_parts) == 0)
1110 nr_minors = nr_parts;
1111
1112 err = xlbd_reserve_minors(minor, nr_minors);
1113 if (err)
1114 return err;
1115
1116 memset(&info->tag_set, 0, sizeof(info->tag_set));
1117 info->tag_set.ops = &blkfront_mq_ops;
1118 info->tag_set.nr_hw_queues = info->nr_rings;
1119 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1120 /*
1121 * When indirect descriptior is not supported, the I/O request
1122 * will be split between multiple request in the ring.
1123 * To avoid problems when sending the request, divide by
1124 * 2 the depth of the queue.
1125 */
1126 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
1127 } else
1128 info->tag_set.queue_depth = BLK_RING_SIZE(info);
1129 info->tag_set.numa_node = NUMA_NO_NODE;
1130 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1131 info->tag_set.cmd_size = sizeof(struct blkif_req);
1132 info->tag_set.driver_data = info;
1133
1134 err = blk_mq_alloc_tag_set(&info->tag_set);
1135 if (err)
1136 goto out_release_minors;
1137
1138 gd = blk_mq_alloc_disk(&info->tag_set, info);
1139 if (IS_ERR(gd)) {
1140 err = PTR_ERR(gd);
1141 goto out_free_tag_set;
1142 }
1143
1144 strcpy(gd->disk_name, DEV_NAME);
1145 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1146 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1147 if (nr_minors > 1)
1148 *ptr = 0;
1149 else
1150 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1151 "%d", minor & (nr_parts - 1));
1152
1153 gd->major = XENVBD_MAJOR;
1154 gd->first_minor = minor;
1155 gd->minors = nr_minors;
1156 gd->fops = &xlvbd_block_fops;
1157 gd->private_data = info;
1158 set_capacity(gd, capacity);
1159
1160 info->rq = gd->queue;
1161 info->gd = gd;
1162 info->sector_size = sector_size;
1163 info->physical_sector_size = physical_sector_size;
1164 blkif_set_queue_limits(info);
1165
1166 xlvbd_flush(info);
1167
1168 if (info->vdisk_info & VDISK_READONLY)
1169 set_disk_ro(gd, 1);
1170 if (info->vdisk_info & VDISK_REMOVABLE)
1171 gd->flags |= GENHD_FL_REMOVABLE;
1172
1173 return 0;
1174
1175out_free_tag_set:
1176 blk_mq_free_tag_set(&info->tag_set);
1177out_release_minors:
1178 xlbd_release_minors(minor, nr_minors);
1179 return err;
1180}
1181
1182/* Already hold rinfo->ring_lock. */
1183static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1184{
1185 if (!RING_FULL(&rinfo->ring))
1186 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1187}
1188
1189static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1190{
1191 unsigned long flags;
1192
1193 spin_lock_irqsave(&rinfo->ring_lock, flags);
1194 kick_pending_request_queues_locked(rinfo);
1195 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1196}
1197
1198static void blkif_restart_queue(struct work_struct *work)
1199{
1200 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1201
1202 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1203 kick_pending_request_queues(rinfo);
1204}
1205
1206static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1207{
1208 struct grant *persistent_gnt, *n;
1209 struct blkfront_info *info = rinfo->dev_info;
1210 int i, j, segs;
1211
1212 /*
1213 * Remove indirect pages, this only happens when using indirect
1214 * descriptors but not persistent grants
1215 */
1216 if (!list_empty(&rinfo->indirect_pages)) {
1217 struct page *indirect_page, *n;
1218
1219 BUG_ON(info->bounce);
1220 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1221 list_del(&indirect_page->lru);
1222 __free_page(indirect_page);
1223 }
1224 }
1225
1226 /* Remove all persistent grants. */
1227 if (!list_empty(&rinfo->grants)) {
1228 list_for_each_entry_safe(persistent_gnt, n,
1229 &rinfo->grants, node) {
1230 list_del(&persistent_gnt->node);
1231 if (persistent_gnt->gref != INVALID_GRANT_REF) {
1232 gnttab_end_foreign_access(persistent_gnt->gref,
1233 NULL);
1234 rinfo->persistent_gnts_c--;
1235 }
1236 if (info->bounce)
1237 __free_page(persistent_gnt->page);
1238 kfree(persistent_gnt);
1239 }
1240 }
1241 BUG_ON(rinfo->persistent_gnts_c != 0);
1242
1243 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1244 /*
1245 * Clear persistent grants present in requests already
1246 * on the shared ring
1247 */
1248 if (!rinfo->shadow[i].request)
1249 goto free_shadow;
1250
1251 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1252 rinfo->shadow[i].req.u.indirect.nr_segments :
1253 rinfo->shadow[i].req.u.rw.nr_segments;
1254 for (j = 0; j < segs; j++) {
1255 persistent_gnt = rinfo->shadow[i].grants_used[j];
1256 gnttab_end_foreign_access(persistent_gnt->gref, NULL);
1257 if (info->bounce)
1258 __free_page(persistent_gnt->page);
1259 kfree(persistent_gnt);
1260 }
1261
1262 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1263 /*
1264 * If this is not an indirect operation don't try to
1265 * free indirect segments
1266 */
1267 goto free_shadow;
1268
1269 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1270 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1271 gnttab_end_foreign_access(persistent_gnt->gref, NULL);
1272 __free_page(persistent_gnt->page);
1273 kfree(persistent_gnt);
1274 }
1275
1276free_shadow:
1277 kvfree(rinfo->shadow[i].grants_used);
1278 rinfo->shadow[i].grants_used = NULL;
1279 kvfree(rinfo->shadow[i].indirect_grants);
1280 rinfo->shadow[i].indirect_grants = NULL;
1281 kvfree(rinfo->shadow[i].sg);
1282 rinfo->shadow[i].sg = NULL;
1283 }
1284
1285 /* No more gnttab callback work. */
1286 gnttab_cancel_free_callback(&rinfo->callback);
1287
1288 /* Flush gnttab callback work. Must be done with no locks held. */
1289 flush_work(&rinfo->work);
1290
1291 /* Free resources associated with old device channel. */
1292 xenbus_teardown_ring((void **)&rinfo->ring.sring, info->nr_ring_pages,
1293 rinfo->ring_ref);
1294
1295 if (rinfo->irq)
1296 unbind_from_irqhandler(rinfo->irq, rinfo);
1297 rinfo->evtchn = rinfo->irq = 0;
1298}
1299
1300static void blkif_free(struct blkfront_info *info, int suspend)
1301{
1302 unsigned int i;
1303 struct blkfront_ring_info *rinfo;
1304
1305 /* Prevent new requests being issued until we fix things up. */
1306 info->connected = suspend ?
1307 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1308 /* No more blkif_request(). */
1309 if (info->rq)
1310 blk_mq_stop_hw_queues(info->rq);
1311
1312 for_each_rinfo(info, rinfo, i)
1313 blkif_free_ring(rinfo);
1314
1315 kvfree(info->rinfo);
1316 info->rinfo = NULL;
1317 info->nr_rings = 0;
1318}
1319
1320struct copy_from_grant {
1321 const struct blk_shadow *s;
1322 unsigned int grant_idx;
1323 unsigned int bvec_offset;
1324 char *bvec_data;
1325};
1326
1327static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1328 unsigned int len, void *data)
1329{
1330 struct copy_from_grant *info = data;
1331 char *shared_data;
1332 /* Convenient aliases */
1333 const struct blk_shadow *s = info->s;
1334
1335 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1336
1337 memcpy(info->bvec_data + info->bvec_offset,
1338 shared_data + offset, len);
1339
1340 info->bvec_offset += len;
1341 info->grant_idx++;
1342
1343 kunmap_atomic(shared_data);
1344}
1345
1346static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1347{
1348 switch (rsp)
1349 {
1350 case BLKIF_RSP_OKAY:
1351 return REQ_DONE;
1352 case BLKIF_RSP_EOPNOTSUPP:
1353 return REQ_EOPNOTSUPP;
1354 case BLKIF_RSP_ERROR:
1355 default:
1356 return REQ_ERROR;
1357 }
1358}
1359
1360/*
1361 * Get the final status of the block request based on two ring response
1362 */
1363static int blkif_get_final_status(enum blk_req_status s1,
1364 enum blk_req_status s2)
1365{
1366 BUG_ON(s1 < REQ_DONE);
1367 BUG_ON(s2 < REQ_DONE);
1368
1369 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1370 return BLKIF_RSP_ERROR;
1371 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1372 return BLKIF_RSP_EOPNOTSUPP;
1373 return BLKIF_RSP_OKAY;
1374}
1375
1376/*
1377 * Return values:
1378 * 1 response processed.
1379 * 0 missing further responses.
1380 * -1 error while processing.
1381 */
1382static int blkif_completion(unsigned long *id,
1383 struct blkfront_ring_info *rinfo,
1384 struct blkif_response *bret)
1385{
1386 int i = 0;
1387 struct scatterlist *sg;
1388 int num_sg, num_grant;
1389 struct blkfront_info *info = rinfo->dev_info;
1390 struct blk_shadow *s = &rinfo->shadow[*id];
1391 struct copy_from_grant data = {
1392 .grant_idx = 0,
1393 };
1394
1395 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1396 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1397
1398 /* The I/O request may be split in two. */
1399 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1400 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1401
1402 /* Keep the status of the current response in shadow. */
1403 s->status = blkif_rsp_to_req_status(bret->status);
1404
1405 /* Wait the second response if not yet here. */
1406 if (s2->status < REQ_DONE)
1407 return 0;
1408
1409 bret->status = blkif_get_final_status(s->status,
1410 s2->status);
1411
1412 /*
1413 * All the grants is stored in the first shadow in order
1414 * to make the completion code simpler.
1415 */
1416 num_grant += s2->req.u.rw.nr_segments;
1417
1418 /*
1419 * The two responses may not come in order. Only the
1420 * first request will store the scatter-gather list.
1421 */
1422 if (s2->num_sg != 0) {
1423 /* Update "id" with the ID of the first response. */
1424 *id = s->associated_id;
1425 s = s2;
1426 }
1427
1428 /*
1429 * We don't need anymore the second request, so recycling
1430 * it now.
1431 */
1432 if (add_id_to_freelist(rinfo, s->associated_id))
1433 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1434 info->gd->disk_name, s->associated_id);
1435 }
1436
1437 data.s = s;
1438 num_sg = s->num_sg;
1439
1440 if (bret->operation == BLKIF_OP_READ && info->bounce) {
1441 for_each_sg(s->sg, sg, num_sg, i) {
1442 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1443
1444 data.bvec_offset = sg->offset;
1445 data.bvec_data = kmap_atomic(sg_page(sg));
1446
1447 gnttab_foreach_grant_in_range(sg_page(sg),
1448 sg->offset,
1449 sg->length,
1450 blkif_copy_from_grant,
1451 &data);
1452
1453 kunmap_atomic(data.bvec_data);
1454 }
1455 }
1456 /* Add the persistent grant into the list of free grants */
1457 for (i = 0; i < num_grant; i++) {
1458 if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) {
1459 /*
1460 * If the grant is still mapped by the backend (the
1461 * backend has chosen to make this grant persistent)
1462 * we add it at the head of the list, so it will be
1463 * reused first.
1464 */
1465 if (!info->feature_persistent) {
1466 pr_alert("backed has not unmapped grant: %u\n",
1467 s->grants_used[i]->gref);
1468 return -1;
1469 }
1470 list_add(&s->grants_used[i]->node, &rinfo->grants);
1471 rinfo->persistent_gnts_c++;
1472 } else {
1473 /*
1474 * If the grant is not mapped by the backend we add it
1475 * to the tail of the list, so it will not be picked
1476 * again unless we run out of persistent grants.
1477 */
1478 s->grants_used[i]->gref = INVALID_GRANT_REF;
1479 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1480 }
1481 }
1482 if (s->req.operation == BLKIF_OP_INDIRECT) {
1483 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1484 if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) {
1485 if (!info->feature_persistent) {
1486 pr_alert("backed has not unmapped grant: %u\n",
1487 s->indirect_grants[i]->gref);
1488 return -1;
1489 }
1490 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1491 rinfo->persistent_gnts_c++;
1492 } else {
1493 struct page *indirect_page;
1494
1495 /*
1496 * Add the used indirect page back to the list of
1497 * available pages for indirect grefs.
1498 */
1499 if (!info->bounce) {
1500 indirect_page = s->indirect_grants[i]->page;
1501 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1502 }
1503 s->indirect_grants[i]->gref = INVALID_GRANT_REF;
1504 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1505 }
1506 }
1507 }
1508
1509 return 1;
1510}
1511
1512static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1513{
1514 struct request *req;
1515 struct blkif_response bret;
1516 RING_IDX i, rp;
1517 unsigned long flags;
1518 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1519 struct blkfront_info *info = rinfo->dev_info;
1520 unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1521
1522 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1523 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1524 return IRQ_HANDLED;
1525 }
1526
1527 spin_lock_irqsave(&rinfo->ring_lock, flags);
1528 again:
1529 rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1530 virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1531 if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1532 pr_alert("%s: illegal number of responses %u\n",
1533 info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1534 goto err;
1535 }
1536
1537 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1538 unsigned long id;
1539 unsigned int op;
1540
1541 eoiflag = 0;
1542
1543 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1544 id = bret.id;
1545
1546 /*
1547 * The backend has messed up and given us an id that we would
1548 * never have given to it (we stamp it up to BLK_RING_SIZE -
1549 * look in get_id_from_freelist.
1550 */
1551 if (id >= BLK_RING_SIZE(info)) {
1552 pr_alert("%s: response has incorrect id (%ld)\n",
1553 info->gd->disk_name, id);
1554 goto err;
1555 }
1556 if (rinfo->shadow[id].status != REQ_WAITING) {
1557 pr_alert("%s: response references no pending request\n",
1558 info->gd->disk_name);
1559 goto err;
1560 }
1561
1562 rinfo->shadow[id].status = REQ_PROCESSING;
1563 req = rinfo->shadow[id].request;
1564
1565 op = rinfo->shadow[id].req.operation;
1566 if (op == BLKIF_OP_INDIRECT)
1567 op = rinfo->shadow[id].req.u.indirect.indirect_op;
1568 if (bret.operation != op) {
1569 pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1570 info->gd->disk_name, bret.operation, op);
1571 goto err;
1572 }
1573
1574 if (bret.operation != BLKIF_OP_DISCARD) {
1575 int ret;
1576
1577 /*
1578 * We may need to wait for an extra response if the
1579 * I/O request is split in 2
1580 */
1581 ret = blkif_completion(&id, rinfo, &bret);
1582 if (!ret)
1583 continue;
1584 if (unlikely(ret < 0))
1585 goto err;
1586 }
1587
1588 if (add_id_to_freelist(rinfo, id)) {
1589 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1590 info->gd->disk_name, op_name(bret.operation), id);
1591 continue;
1592 }
1593
1594 if (bret.status == BLKIF_RSP_OKAY)
1595 blkif_req(req)->error = BLK_STS_OK;
1596 else
1597 blkif_req(req)->error = BLK_STS_IOERR;
1598
1599 switch (bret.operation) {
1600 case BLKIF_OP_DISCARD:
1601 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1602 struct request_queue *rq = info->rq;
1603
1604 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1605 info->gd->disk_name, op_name(bret.operation));
1606 blkif_req(req)->error = BLK_STS_NOTSUPP;
1607 info->feature_discard = 0;
1608 info->feature_secdiscard = 0;
1609 blk_queue_max_discard_sectors(rq, 0);
1610 blk_queue_max_secure_erase_sectors(rq, 0);
1611 }
1612 break;
1613 case BLKIF_OP_FLUSH_DISKCACHE:
1614 case BLKIF_OP_WRITE_BARRIER:
1615 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1616 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1617 info->gd->disk_name, op_name(bret.operation));
1618 blkif_req(req)->error = BLK_STS_NOTSUPP;
1619 }
1620 if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1621 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1622 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1623 info->gd->disk_name, op_name(bret.operation));
1624 blkif_req(req)->error = BLK_STS_NOTSUPP;
1625 }
1626 if (unlikely(blkif_req(req)->error)) {
1627 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1628 blkif_req(req)->error = BLK_STS_OK;
1629 info->feature_fua = 0;
1630 info->feature_flush = 0;
1631 xlvbd_flush(info);
1632 }
1633 fallthrough;
1634 case BLKIF_OP_READ:
1635 case BLKIF_OP_WRITE:
1636 if (unlikely(bret.status != BLKIF_RSP_OKAY))
1637 dev_dbg_ratelimited(&info->xbdev->dev,
1638 "Bad return from blkdev data request: %#x\n",
1639 bret.status);
1640
1641 break;
1642 default:
1643 BUG();
1644 }
1645
1646 if (likely(!blk_should_fake_timeout(req->q)))
1647 blk_mq_complete_request(req);
1648 }
1649
1650 rinfo->ring.rsp_cons = i;
1651
1652 if (i != rinfo->ring.req_prod_pvt) {
1653 int more_to_do;
1654 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1655 if (more_to_do)
1656 goto again;
1657 } else
1658 rinfo->ring.sring->rsp_event = i + 1;
1659
1660 kick_pending_request_queues_locked(rinfo);
1661
1662 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1663
1664 xen_irq_lateeoi(irq, eoiflag);
1665
1666 return IRQ_HANDLED;
1667
1668 err:
1669 info->connected = BLKIF_STATE_ERROR;
1670
1671 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1672
1673 /* No EOI in order to avoid further interrupts. */
1674
1675 pr_alert("%s disabled for further use\n", info->gd->disk_name);
1676 return IRQ_HANDLED;
1677}
1678
1679
1680static int setup_blkring(struct xenbus_device *dev,
1681 struct blkfront_ring_info *rinfo)
1682{
1683 struct blkif_sring *sring;
1684 int err;
1685 struct blkfront_info *info = rinfo->dev_info;
1686 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1687
1688 err = xenbus_setup_ring(dev, GFP_NOIO, (void **)&sring,
1689 info->nr_ring_pages, rinfo->ring_ref);
1690 if (err)
1691 goto fail;
1692
1693 XEN_FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1694
1695 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1696 if (err)
1697 goto fail;
1698
1699 err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1700 0, "blkif", rinfo);
1701 if (err <= 0) {
1702 xenbus_dev_fatal(dev, err,
1703 "bind_evtchn_to_irqhandler failed");
1704 goto fail;
1705 }
1706 rinfo->irq = err;
1707
1708 return 0;
1709fail:
1710 blkif_free(info, 0);
1711 return err;
1712}
1713
1714/*
1715 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1716 * ring buffer may have multi pages depending on ->nr_ring_pages.
1717 */
1718static int write_per_ring_nodes(struct xenbus_transaction xbt,
1719 struct blkfront_ring_info *rinfo, const char *dir)
1720{
1721 int err;
1722 unsigned int i;
1723 const char *message = NULL;
1724 struct blkfront_info *info = rinfo->dev_info;
1725
1726 if (info->nr_ring_pages == 1) {
1727 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1728 if (err) {
1729 message = "writing ring-ref";
1730 goto abort_transaction;
1731 }
1732 } else {
1733 for (i = 0; i < info->nr_ring_pages; i++) {
1734 char ring_ref_name[RINGREF_NAME_LEN];
1735
1736 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1737 err = xenbus_printf(xbt, dir, ring_ref_name,
1738 "%u", rinfo->ring_ref[i]);
1739 if (err) {
1740 message = "writing ring-ref";
1741 goto abort_transaction;
1742 }
1743 }
1744 }
1745
1746 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1747 if (err) {
1748 message = "writing event-channel";
1749 goto abort_transaction;
1750 }
1751
1752 return 0;
1753
1754abort_transaction:
1755 xenbus_transaction_end(xbt, 1);
1756 if (message)
1757 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1758
1759 return err;
1760}
1761
1762/* Enable the persistent grants feature. */
1763static bool feature_persistent = true;
1764module_param(feature_persistent, bool, 0644);
1765MODULE_PARM_DESC(feature_persistent,
1766 "Enables the persistent grants feature");
1767
1768/* Common code used when first setting up, and when resuming. */
1769static int talk_to_blkback(struct xenbus_device *dev,
1770 struct blkfront_info *info)
1771{
1772 const char *message = NULL;
1773 struct xenbus_transaction xbt;
1774 int err;
1775 unsigned int i, max_page_order;
1776 unsigned int ring_page_order;
1777 struct blkfront_ring_info *rinfo;
1778
1779 if (!info)
1780 return -ENODEV;
1781
1782 /* Check if backend is trusted. */
1783 info->bounce = !xen_blkif_trusted ||
1784 !xenbus_read_unsigned(dev->nodename, "trusted", 1);
1785
1786 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1787 "max-ring-page-order", 0);
1788 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1789 info->nr_ring_pages = 1 << ring_page_order;
1790
1791 err = negotiate_mq(info);
1792 if (err)
1793 goto destroy_blkring;
1794
1795 for_each_rinfo(info, rinfo, i) {
1796 /* Create shared ring, alloc event channel. */
1797 err = setup_blkring(dev, rinfo);
1798 if (err)
1799 goto destroy_blkring;
1800 }
1801
1802again:
1803 err = xenbus_transaction_start(&xbt);
1804 if (err) {
1805 xenbus_dev_fatal(dev, err, "starting transaction");
1806 goto destroy_blkring;
1807 }
1808
1809 if (info->nr_ring_pages > 1) {
1810 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1811 ring_page_order);
1812 if (err) {
1813 message = "writing ring-page-order";
1814 goto abort_transaction;
1815 }
1816 }
1817
1818 /* We already got the number of queues/rings in _probe */
1819 if (info->nr_rings == 1) {
1820 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1821 if (err)
1822 goto destroy_blkring;
1823 } else {
1824 char *path;
1825 size_t pathsize;
1826
1827 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1828 info->nr_rings);
1829 if (err) {
1830 message = "writing multi-queue-num-queues";
1831 goto abort_transaction;
1832 }
1833
1834 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1835 path = kmalloc(pathsize, GFP_KERNEL);
1836 if (!path) {
1837 err = -ENOMEM;
1838 message = "ENOMEM while writing ring references";
1839 goto abort_transaction;
1840 }
1841
1842 for_each_rinfo(info, rinfo, i) {
1843 memset(path, 0, pathsize);
1844 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1845 err = write_per_ring_nodes(xbt, rinfo, path);
1846 if (err) {
1847 kfree(path);
1848 goto destroy_blkring;
1849 }
1850 }
1851 kfree(path);
1852 }
1853 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1854 XEN_IO_PROTO_ABI_NATIVE);
1855 if (err) {
1856 message = "writing protocol";
1857 goto abort_transaction;
1858 }
1859 info->feature_persistent_parm = feature_persistent;
1860 err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1861 info->feature_persistent_parm);
1862 if (err)
1863 dev_warn(&dev->dev,
1864 "writing persistent grants feature to xenbus");
1865
1866 err = xenbus_transaction_end(xbt, 0);
1867 if (err) {
1868 if (err == -EAGAIN)
1869 goto again;
1870 xenbus_dev_fatal(dev, err, "completing transaction");
1871 goto destroy_blkring;
1872 }
1873
1874 for_each_rinfo(info, rinfo, i) {
1875 unsigned int j;
1876
1877 for (j = 0; j < BLK_RING_SIZE(info); j++)
1878 rinfo->shadow[j].req.u.rw.id = j + 1;
1879 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1880 }
1881 xenbus_switch_state(dev, XenbusStateInitialised);
1882
1883 return 0;
1884
1885 abort_transaction:
1886 xenbus_transaction_end(xbt, 1);
1887 if (message)
1888 xenbus_dev_fatal(dev, err, "%s", message);
1889 destroy_blkring:
1890 blkif_free(info, 0);
1891 return err;
1892}
1893
1894static int negotiate_mq(struct blkfront_info *info)
1895{
1896 unsigned int backend_max_queues;
1897 unsigned int i;
1898 struct blkfront_ring_info *rinfo;
1899
1900 BUG_ON(info->nr_rings);
1901
1902 /* Check if backend supports multiple queues. */
1903 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1904 "multi-queue-max-queues", 1);
1905 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1906 /* We need at least one ring. */
1907 if (!info->nr_rings)
1908 info->nr_rings = 1;
1909
1910 info->rinfo_size = struct_size(info->rinfo, shadow,
1911 BLK_RING_SIZE(info));
1912 info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1913 if (!info->rinfo) {
1914 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1915 info->nr_rings = 0;
1916 return -ENOMEM;
1917 }
1918
1919 for_each_rinfo(info, rinfo, i) {
1920 INIT_LIST_HEAD(&rinfo->indirect_pages);
1921 INIT_LIST_HEAD(&rinfo->grants);
1922 rinfo->dev_info = info;
1923 INIT_WORK(&rinfo->work, blkif_restart_queue);
1924 spin_lock_init(&rinfo->ring_lock);
1925 }
1926 return 0;
1927}
1928
1929/*
1930 * Entry point to this code when a new device is created. Allocate the basic
1931 * structures and the ring buffer for communication with the backend, and
1932 * inform the backend of the appropriate details for those. Switch to
1933 * Initialised state.
1934 */
1935static int blkfront_probe(struct xenbus_device *dev,
1936 const struct xenbus_device_id *id)
1937{
1938 int err, vdevice;
1939 struct blkfront_info *info;
1940
1941 /* FIXME: Use dynamic device id if this is not set. */
1942 err = xenbus_scanf(XBT_NIL, dev->nodename,
1943 "virtual-device", "%i", &vdevice);
1944 if (err != 1) {
1945 /* go looking in the extended area instead */
1946 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1947 "%i", &vdevice);
1948 if (err != 1) {
1949 xenbus_dev_fatal(dev, err, "reading virtual-device");
1950 return err;
1951 }
1952 }
1953
1954 if (xen_hvm_domain()) {
1955 char *type;
1956 int len;
1957 /* no unplug has been done: do not hook devices != xen vbds */
1958 if (xen_has_pv_and_legacy_disk_devices()) {
1959 int major;
1960
1961 if (!VDEV_IS_EXTENDED(vdevice))
1962 major = BLKIF_MAJOR(vdevice);
1963 else
1964 major = XENVBD_MAJOR;
1965
1966 if (major != XENVBD_MAJOR) {
1967 printk(KERN_INFO
1968 "%s: HVM does not support vbd %d as xen block device\n",
1969 __func__, vdevice);
1970 return -ENODEV;
1971 }
1972 }
1973 /* do not create a PV cdrom device if we are an HVM guest */
1974 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1975 if (IS_ERR(type))
1976 return -ENODEV;
1977 if (strncmp(type, "cdrom", 5) == 0) {
1978 kfree(type);
1979 return -ENODEV;
1980 }
1981 kfree(type);
1982 }
1983 info = kzalloc(sizeof(*info), GFP_KERNEL);
1984 if (!info) {
1985 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1986 return -ENOMEM;
1987 }
1988
1989 info->xbdev = dev;
1990
1991 mutex_init(&info->mutex);
1992 info->vdevice = vdevice;
1993 info->connected = BLKIF_STATE_DISCONNECTED;
1994
1995 /* Front end dir is a number, which is used as the id. */
1996 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1997 dev_set_drvdata(&dev->dev, info);
1998
1999 mutex_lock(&blkfront_mutex);
2000 list_add(&info->info_list, &info_list);
2001 mutex_unlock(&blkfront_mutex);
2002
2003 return 0;
2004}
2005
2006static int blkif_recover(struct blkfront_info *info)
2007{
2008 unsigned int r_index;
2009 struct request *req, *n;
2010 int rc;
2011 struct bio *bio;
2012 unsigned int segs;
2013 struct blkfront_ring_info *rinfo;
2014
2015 blkfront_gather_backend_features(info);
2016 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2017 blkif_set_queue_limits(info);
2018 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2019 blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2020
2021 for_each_rinfo(info, rinfo, r_index) {
2022 rc = blkfront_setup_indirect(rinfo);
2023 if (rc)
2024 return rc;
2025 }
2026 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2027
2028 /* Now safe for us to use the shared ring */
2029 info->connected = BLKIF_STATE_CONNECTED;
2030
2031 for_each_rinfo(info, rinfo, r_index) {
2032 /* Kick any other new requests queued since we resumed */
2033 kick_pending_request_queues(rinfo);
2034 }
2035
2036 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2037 /* Requeue pending requests (flush or discard) */
2038 list_del_init(&req->queuelist);
2039 BUG_ON(req->nr_phys_segments > segs);
2040 blk_mq_requeue_request(req, false);
2041 }
2042 blk_mq_start_stopped_hw_queues(info->rq, true);
2043 blk_mq_kick_requeue_list(info->rq);
2044
2045 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2046 /* Traverse the list of pending bios and re-queue them */
2047 submit_bio(bio);
2048 }
2049
2050 return 0;
2051}
2052
2053/*
2054 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2055 * driver restart. We tear down our blkif structure and recreate it, but
2056 * leave the device-layer structures intact so that this is transparent to the
2057 * rest of the kernel.
2058 */
2059static int blkfront_resume(struct xenbus_device *dev)
2060{
2061 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2062 int err = 0;
2063 unsigned int i, j;
2064 struct blkfront_ring_info *rinfo;
2065
2066 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2067
2068 bio_list_init(&info->bio_list);
2069 INIT_LIST_HEAD(&info->requests);
2070 for_each_rinfo(info, rinfo, i) {
2071 struct bio_list merge_bio;
2072 struct blk_shadow *shadow = rinfo->shadow;
2073
2074 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2075 /* Not in use? */
2076 if (!shadow[j].request)
2077 continue;
2078
2079 /*
2080 * Get the bios in the request so we can re-queue them.
2081 */
2082 if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2083 req_op(shadow[j].request) == REQ_OP_DISCARD ||
2084 req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2085 shadow[j].request->cmd_flags & REQ_FUA) {
2086 /*
2087 * Flush operations don't contain bios, so
2088 * we need to requeue the whole request
2089 *
2090 * XXX: but this doesn't make any sense for a
2091 * write with the FUA flag set..
2092 */
2093 list_add(&shadow[j].request->queuelist, &info->requests);
2094 continue;
2095 }
2096 merge_bio.head = shadow[j].request->bio;
2097 merge_bio.tail = shadow[j].request->biotail;
2098 bio_list_merge(&info->bio_list, &merge_bio);
2099 shadow[j].request->bio = NULL;
2100 blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2101 }
2102 }
2103
2104 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2105
2106 err = talk_to_blkback(dev, info);
2107 if (!err)
2108 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2109
2110 /*
2111 * We have to wait for the backend to switch to
2112 * connected state, since we want to read which
2113 * features it supports.
2114 */
2115
2116 return err;
2117}
2118
2119static void blkfront_closing(struct blkfront_info *info)
2120{
2121 struct xenbus_device *xbdev = info->xbdev;
2122 struct blkfront_ring_info *rinfo;
2123 unsigned int i;
2124
2125 if (xbdev->state == XenbusStateClosing)
2126 return;
2127
2128 /* No more blkif_request(). */
2129 if (info->rq && info->gd) {
2130 blk_mq_stop_hw_queues(info->rq);
2131 blk_mark_disk_dead(info->gd);
2132 }
2133
2134 for_each_rinfo(info, rinfo, i) {
2135 /* No more gnttab callback work. */
2136 gnttab_cancel_free_callback(&rinfo->callback);
2137
2138 /* Flush gnttab callback work. Must be done with no locks held. */
2139 flush_work(&rinfo->work);
2140 }
2141
2142 xenbus_frontend_closed(xbdev);
2143}
2144
2145static void blkfront_setup_discard(struct blkfront_info *info)
2146{
2147 info->feature_discard = 1;
2148 info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2149 "discard-granularity",
2150 0);
2151 info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2152 "discard-alignment", 0);
2153 info->feature_secdiscard =
2154 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2155 0);
2156}
2157
2158static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2159{
2160 unsigned int psegs, grants, memflags;
2161 int err, i;
2162 struct blkfront_info *info = rinfo->dev_info;
2163
2164 memflags = memalloc_noio_save();
2165
2166 if (info->max_indirect_segments == 0) {
2167 if (!HAS_EXTRA_REQ)
2168 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2169 else {
2170 /*
2171 * When an extra req is required, the maximum
2172 * grants supported is related to the size of the
2173 * Linux block segment.
2174 */
2175 grants = GRANTS_PER_PSEG;
2176 }
2177 }
2178 else
2179 grants = info->max_indirect_segments;
2180 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2181
2182 err = fill_grant_buffer(rinfo,
2183 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2184 if (err)
2185 goto out_of_memory;
2186
2187 if (!info->bounce && info->max_indirect_segments) {
2188 /*
2189 * We are using indirect descriptors but don't have a bounce
2190 * buffer, we need to allocate a set of pages that can be
2191 * used for mapping indirect grefs
2192 */
2193 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2194
2195 BUG_ON(!list_empty(&rinfo->indirect_pages));
2196 for (i = 0; i < num; i++) {
2197 struct page *indirect_page = alloc_page(GFP_KERNEL |
2198 __GFP_ZERO);
2199 if (!indirect_page)
2200 goto out_of_memory;
2201 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2202 }
2203 }
2204
2205 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2206 rinfo->shadow[i].grants_used =
2207 kvcalloc(grants,
2208 sizeof(rinfo->shadow[i].grants_used[0]),
2209 GFP_KERNEL);
2210 rinfo->shadow[i].sg = kvcalloc(psegs,
2211 sizeof(rinfo->shadow[i].sg[0]),
2212 GFP_KERNEL);
2213 if (info->max_indirect_segments)
2214 rinfo->shadow[i].indirect_grants =
2215 kvcalloc(INDIRECT_GREFS(grants),
2216 sizeof(rinfo->shadow[i].indirect_grants[0]),
2217 GFP_KERNEL);
2218 if ((rinfo->shadow[i].grants_used == NULL) ||
2219 (rinfo->shadow[i].sg == NULL) ||
2220 (info->max_indirect_segments &&
2221 (rinfo->shadow[i].indirect_grants == NULL)))
2222 goto out_of_memory;
2223 sg_init_table(rinfo->shadow[i].sg, psegs);
2224 }
2225
2226 memalloc_noio_restore(memflags);
2227
2228 return 0;
2229
2230out_of_memory:
2231 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2232 kvfree(rinfo->shadow[i].grants_used);
2233 rinfo->shadow[i].grants_used = NULL;
2234 kvfree(rinfo->shadow[i].sg);
2235 rinfo->shadow[i].sg = NULL;
2236 kvfree(rinfo->shadow[i].indirect_grants);
2237 rinfo->shadow[i].indirect_grants = NULL;
2238 }
2239 if (!list_empty(&rinfo->indirect_pages)) {
2240 struct page *indirect_page, *n;
2241 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2242 list_del(&indirect_page->lru);
2243 __free_page(indirect_page);
2244 }
2245 }
2246
2247 memalloc_noio_restore(memflags);
2248
2249 return -ENOMEM;
2250}
2251
2252/*
2253 * Gather all backend feature-*
2254 */
2255static void blkfront_gather_backend_features(struct blkfront_info *info)
2256{
2257 unsigned int indirect_segments;
2258
2259 info->feature_flush = 0;
2260 info->feature_fua = 0;
2261
2262 /*
2263 * If there's no "feature-barrier" defined, then it means
2264 * we're dealing with a very old backend which writes
2265 * synchronously; nothing to do.
2266 *
2267 * If there are barriers, then we use flush.
2268 */
2269 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2270 info->feature_flush = 1;
2271 info->feature_fua = 1;
2272 }
2273
2274 /*
2275 * And if there is "feature-flush-cache" use that above
2276 * barriers.
2277 */
2278 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2279 0)) {
2280 info->feature_flush = 1;
2281 info->feature_fua = 0;
2282 }
2283
2284 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2285 blkfront_setup_discard(info);
2286
2287 if (info->feature_persistent_parm)
2288 info->feature_persistent =
2289 !!xenbus_read_unsigned(info->xbdev->otherend,
2290 "feature-persistent", 0);
2291 if (info->feature_persistent)
2292 info->bounce = true;
2293
2294 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2295 "feature-max-indirect-segments", 0);
2296 if (indirect_segments > xen_blkif_max_segments)
2297 indirect_segments = xen_blkif_max_segments;
2298 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2299 indirect_segments = 0;
2300 info->max_indirect_segments = indirect_segments;
2301
2302 if (info->feature_persistent) {
2303 mutex_lock(&blkfront_mutex);
2304 schedule_delayed_work(&blkfront_work, HZ * 10);
2305 mutex_unlock(&blkfront_mutex);
2306 }
2307}
2308
2309/*
2310 * Invoked when the backend is finally 'ready' (and has told produced
2311 * the details about the physical device - #sectors, size, etc).
2312 */
2313static void blkfront_connect(struct blkfront_info *info)
2314{
2315 unsigned long long sectors;
2316 unsigned long sector_size;
2317 unsigned int physical_sector_size;
2318 int err, i;
2319 struct blkfront_ring_info *rinfo;
2320
2321 switch (info->connected) {
2322 case BLKIF_STATE_CONNECTED:
2323 /*
2324 * Potentially, the back-end may be signalling
2325 * a capacity change; update the capacity.
2326 */
2327 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2328 "sectors", "%Lu", §ors);
2329 if (XENBUS_EXIST_ERR(err))
2330 return;
2331 printk(KERN_INFO "Setting capacity to %Lu\n",
2332 sectors);
2333 set_capacity_and_notify(info->gd, sectors);
2334
2335 return;
2336 case BLKIF_STATE_SUSPENDED:
2337 /*
2338 * If we are recovering from suspension, we need to wait
2339 * for the backend to announce it's features before
2340 * reconnecting, at least we need to know if the backend
2341 * supports indirect descriptors, and how many.
2342 */
2343 blkif_recover(info);
2344 return;
2345
2346 default:
2347 break;
2348 }
2349
2350 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2351 __func__, info->xbdev->otherend);
2352
2353 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2354 "sectors", "%llu", §ors,
2355 "info", "%u", &info->vdisk_info,
2356 "sector-size", "%lu", §or_size,
2357 NULL);
2358 if (err) {
2359 xenbus_dev_fatal(info->xbdev, err,
2360 "reading backend fields at %s",
2361 info->xbdev->otherend);
2362 return;
2363 }
2364
2365 /*
2366 * physical-sector-size is a newer field, so old backends may not
2367 * provide this. Assume physical sector size to be the same as
2368 * sector_size in that case.
2369 */
2370 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2371 "physical-sector-size",
2372 sector_size);
2373 blkfront_gather_backend_features(info);
2374 for_each_rinfo(info, rinfo, i) {
2375 err = blkfront_setup_indirect(rinfo);
2376 if (err) {
2377 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2378 info->xbdev->otherend);
2379 blkif_free(info, 0);
2380 break;
2381 }
2382 }
2383
2384 err = xlvbd_alloc_gendisk(sectors, info, sector_size,
2385 physical_sector_size);
2386 if (err) {
2387 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2388 info->xbdev->otherend);
2389 goto fail;
2390 }
2391
2392 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2393
2394 /* Kick pending requests. */
2395 info->connected = BLKIF_STATE_CONNECTED;
2396 for_each_rinfo(info, rinfo, i)
2397 kick_pending_request_queues(rinfo);
2398
2399 err = device_add_disk(&info->xbdev->dev, info->gd, NULL);
2400 if (err) {
2401 put_disk(info->gd);
2402 blk_mq_free_tag_set(&info->tag_set);
2403 info->rq = NULL;
2404 goto fail;
2405 }
2406
2407 info->is_ready = 1;
2408 return;
2409
2410fail:
2411 blkif_free(info, 0);
2412 return;
2413}
2414
2415/*
2416 * Callback received when the backend's state changes.
2417 */
2418static void blkback_changed(struct xenbus_device *dev,
2419 enum xenbus_state backend_state)
2420{
2421 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2422
2423 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2424
2425 switch (backend_state) {
2426 case XenbusStateInitWait:
2427 if (dev->state != XenbusStateInitialising)
2428 break;
2429 if (talk_to_blkback(dev, info))
2430 break;
2431 break;
2432 case XenbusStateInitialising:
2433 case XenbusStateInitialised:
2434 case XenbusStateReconfiguring:
2435 case XenbusStateReconfigured:
2436 case XenbusStateUnknown:
2437 break;
2438
2439 case XenbusStateConnected:
2440 /*
2441 * talk_to_blkback sets state to XenbusStateInitialised
2442 * and blkfront_connect sets it to XenbusStateConnected
2443 * (if connection went OK).
2444 *
2445 * If the backend (or toolstack) decides to poke at backend
2446 * state (and re-trigger the watch by setting the state repeatedly
2447 * to XenbusStateConnected (4)) we need to deal with this.
2448 * This is allowed as this is used to communicate to the guest
2449 * that the size of disk has changed!
2450 */
2451 if ((dev->state != XenbusStateInitialised) &&
2452 (dev->state != XenbusStateConnected)) {
2453 if (talk_to_blkback(dev, info))
2454 break;
2455 }
2456
2457 blkfront_connect(info);
2458 break;
2459
2460 case XenbusStateClosed:
2461 if (dev->state == XenbusStateClosed)
2462 break;
2463 fallthrough;
2464 case XenbusStateClosing:
2465 blkfront_closing(info);
2466 break;
2467 }
2468}
2469
2470static void blkfront_remove(struct xenbus_device *xbdev)
2471{
2472 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2473
2474 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2475
2476 if (info->gd)
2477 del_gendisk(info->gd);
2478
2479 mutex_lock(&blkfront_mutex);
2480 list_del(&info->info_list);
2481 mutex_unlock(&blkfront_mutex);
2482
2483 blkif_free(info, 0);
2484 if (info->gd) {
2485 xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2486 put_disk(info->gd);
2487 blk_mq_free_tag_set(&info->tag_set);
2488 }
2489
2490 kfree(info);
2491}
2492
2493static int blkfront_is_ready(struct xenbus_device *dev)
2494{
2495 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2496
2497 return info->is_ready && info->xbdev;
2498}
2499
2500static const struct block_device_operations xlvbd_block_fops =
2501{
2502 .owner = THIS_MODULE,
2503 .getgeo = blkif_getgeo,
2504 .ioctl = blkif_ioctl,
2505 .compat_ioctl = blkdev_compat_ptr_ioctl,
2506};
2507
2508
2509static const struct xenbus_device_id blkfront_ids[] = {
2510 { "vbd" },
2511 { "" }
2512};
2513
2514static struct xenbus_driver blkfront_driver = {
2515 .ids = blkfront_ids,
2516 .probe = blkfront_probe,
2517 .remove = blkfront_remove,
2518 .resume = blkfront_resume,
2519 .otherend_changed = blkback_changed,
2520 .is_ready = blkfront_is_ready,
2521};
2522
2523static void purge_persistent_grants(struct blkfront_info *info)
2524{
2525 unsigned int i;
2526 unsigned long flags;
2527 struct blkfront_ring_info *rinfo;
2528
2529 for_each_rinfo(info, rinfo, i) {
2530 struct grant *gnt_list_entry, *tmp;
2531 LIST_HEAD(grants);
2532
2533 spin_lock_irqsave(&rinfo->ring_lock, flags);
2534
2535 if (rinfo->persistent_gnts_c == 0) {
2536 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2537 continue;
2538 }
2539
2540 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2541 node) {
2542 if (gnt_list_entry->gref == INVALID_GRANT_REF ||
2543 !gnttab_try_end_foreign_access(gnt_list_entry->gref))
2544 continue;
2545
2546 list_del(&gnt_list_entry->node);
2547 rinfo->persistent_gnts_c--;
2548 gnt_list_entry->gref = INVALID_GRANT_REF;
2549 list_add_tail(&gnt_list_entry->node, &grants);
2550 }
2551
2552 list_splice_tail(&grants, &rinfo->grants);
2553
2554 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2555 }
2556}
2557
2558static void blkfront_delay_work(struct work_struct *work)
2559{
2560 struct blkfront_info *info;
2561 bool need_schedule_work = false;
2562
2563 /*
2564 * Note that when using bounce buffers but not persistent grants
2565 * there's no need to run blkfront_delay_work because grants are
2566 * revoked in blkif_completion or else an error is reported and the
2567 * connection is closed.
2568 */
2569
2570 mutex_lock(&blkfront_mutex);
2571
2572 list_for_each_entry(info, &info_list, info_list) {
2573 if (info->feature_persistent) {
2574 need_schedule_work = true;
2575 mutex_lock(&info->mutex);
2576 purge_persistent_grants(info);
2577 mutex_unlock(&info->mutex);
2578 }
2579 }
2580
2581 if (need_schedule_work)
2582 schedule_delayed_work(&blkfront_work, HZ * 10);
2583
2584 mutex_unlock(&blkfront_mutex);
2585}
2586
2587static int __init xlblk_init(void)
2588{
2589 int ret;
2590 int nr_cpus = num_online_cpus();
2591
2592 if (!xen_domain())
2593 return -ENODEV;
2594
2595 if (!xen_has_pv_disk_devices())
2596 return -ENODEV;
2597
2598 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2599 pr_warn("xen_blk: can't get major %d with name %s\n",
2600 XENVBD_MAJOR, DEV_NAME);
2601 return -ENODEV;
2602 }
2603
2604 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2605 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2606
2607 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2608 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2609 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2610 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2611 }
2612
2613 if (xen_blkif_max_queues > nr_cpus) {
2614 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2615 xen_blkif_max_queues, nr_cpus);
2616 xen_blkif_max_queues = nr_cpus;
2617 }
2618
2619 INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2620
2621 ret = xenbus_register_frontend(&blkfront_driver);
2622 if (ret) {
2623 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2624 return ret;
2625 }
2626
2627 return 0;
2628}
2629module_init(xlblk_init);
2630
2631
2632static void __exit xlblk_exit(void)
2633{
2634 cancel_delayed_work_sync(&blkfront_work);
2635
2636 xenbus_unregister_driver(&blkfront_driver);
2637 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2638 kfree(minors);
2639}
2640module_exit(xlblk_exit);
2641
2642MODULE_DESCRIPTION("Xen virtual block device frontend");
2643MODULE_LICENSE("GPL");
2644MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2645MODULE_ALIAS("xen:vbd");
2646MODULE_ALIAS("xenblk");